Anthropic is putting its most powerful — and dangerous — model in everyone’s hands, but it’s doing it with guardrails and hard safety limits.

On Tuesday, the AI firm launched Claude Fable 5, the first publicly available version of its Mythos model.

Until yesterday, Mythos, Anthropic’s exploit-finding software, was the model the White House wanted kept from China, capped at roughly 200 vetted security organizations.

Fable is Latin for what mythos is in Greek. It’s live for everyone, and it’s the most capable model Anthropic has ever shipped to the public.

Claude Fable 5 is the same as Mythos, except for the safeguards.

In high-risk areas like cybersecurity, biology, chemistry, and distillation, the model blocks responses and falls back to Claude Opus 4.8.

It means that the user asks about cybersecurity, biology, or chemistry, and Claude Opus 4.8 answers instead.

Fable is included on Pro, Max, Team, and Enterprise seats through June 22. On June 23, it comes off those plans and runs on usage credits until capacity catches up.

This Fable, built on Mythos tech, is available for $10 per million input tokens and $50 per million output tokens, double the price of Opus 4.8. That price alone might serve as a deterrent for widespread use. It is available on paid Claude plans until June 22.

A new policy comes with it. Anthropic will keep all Mythos-class traffic for 30 days, including from business customers, to detect jailbreaks. Anthropic says the data won’t be used to train new models.

Fable 5 / Mythos needs less guidance than older models and is already doing incredible things. For example, Stripe pointed Fable at a 50-million-line codebase and got a migration done in a day that would have taken a full team over two months by hand.

Launched as a preview in April, Mythos was initially limited to a handful of partners due to cybersecurity concerns. Last week, Anthropic expanded access to hundreds of organizations across 15 countries, again focusing on organizations that manage critical infrastructure.

Now, a version of that technology is available to anyone through Anthropic’s Claude API and consumption-based Enterprise plans.

Anthropic is also deploying a new version of Mythos, called Mythos 5, to organizations already approved to access the advanced model.

In addition, Anthropic said that it ran an internal bug bounty that has produced no universal jailbreaks over 1,000 hours of testing. “We then worked with external red-teaming orgs, which also failed to find universal jailbreaks,” said the company.

Wary of what a Mythos-class model could do in the wrong hands, Anthropic says it stress-tested its classifiers with jailbreak attempts before releasing Fable 5.

With the launch of Fable 5 and Mythos 5, Anthropic said it will require a 30-day retention on all traffic, even if enterprises previously had zero-retention agreements. The company said it won’t use the data for training and will use it only to “defend against complex and novel attacks, including new jailbreaks,” and “identify and reduce false positives.”

The policy could set an industry precedent in which access to increasingly powerful models comes with mandatory data-retention policies framed as a safety measure.

Vibe-coding platform Base44 said that Fable is better at “one-shotting full apps” and has excellent tool-calling.

Genspark said Fable outperformed every other model in its evaluations and performed significantly better on tasks such as UI design and game coding.

Rakuten said, “What makes highly autonomous operations possible, the extra thinking pays for itself.”

Fable’s launch comes as Anthropic prepares to enter the public markets, alongside OpenAI and Elon Musk’s SpaceX. Anthropic said it expects demand for Fable 5 to be very high.

NVIDIA unveiled the new open-source Nemotron 3 Ultra this week, a 550 billion-parameter mixture-of-experts AI model for enterprise workflows, coding, and research, with “up to 5x faster inference and up to 30% lower cost than open frontier models in its class,” according to the company.

The chip company said the model will be released on June 4 on Hugging Face, ModelScope, OpenRouter, and build.nvidia.com as NVIDIA NIM microservices, as well as through a broad ecosystem of NVIDIA Cloud Partners, inference platforms, and cloud service providers.

The verified NVIDIA agent skills are available in the Claude Code plug-in marketplace and the Hermes Skills Hub. NVIDIA also released a major collection of open-source physical AI libraries, skills, models, and frameworks, enabling AI agents and developers to stand up workflows that accelerate the development of robotics, autonomous vehicles, and industrial systems.

The Nemotron 3 Ultra models work with several orchestration frameworks for deploying and coordinating agents, including Hermes Agent, LangChain Deep Agents, OpenClaw, OpenHands, and OpenCode.

These new models and datasets for always-on agents are developed in collaboration with the NVIDIA Nemotron Coalition.

In the Artificial Intelligence ranking, Nemotron 3 Ultra scores 48 points, well ahead of other open U.S. models such as Gemma 4 31B (39), Nemotron 3 Super (36), and gpt-oss-120b (33). It doesn’t reach the top open models from China, though. Kimi K2.6 scores 54 points there. The current strongest closed model, Opus 4.8, hits 61 points.

On provider DeepInfra, Nemotron 3 Ultra also delivers more than 300 tokens per second, according to Artificial Analysis. Comparably sized models from DeepSeek or Moonshot currently manage only 50 to 100.

Firms like CrowdStrike and Palantir also use these kinds of agents to process complex data, coordinate tasks, and streamline operations across cybersecurity and enterprise environments.

• “CrowdStrike is using NVIDIA Nemotron models for its specialized agents that continuously identify, prioritize, and remediate vulnerabilities and policy misconfigurations, helping stop adversaries faster while reducing the operational burden on security teams.”
• “Palantir is integrating NVIDIA Nemotron models into its AI FDE (Forward Deployed Engineer) platform to autonomously execute complex tasks, enabling continuous learning from agent interactions to build domain-specific, air-gapped enterprise systems.”

Autonomous agents that write code, generate sub-agents, and remember context across sessions can access local files, learn new tools, and execute advanced workflows with increasing independence. The more capable agents become, the more important it is to have necessary guardrails for the agents to operate within. The critical layer is a runtime with adjustable privacy and security controls that make autonomous agents safer to deploy at scale.

Canonical will integrate OpenShell with Ubuntu via supported snaps and rocks (aka OCI-compliant containers) to run autonomous agents on enterprise servers worldwide.

Red Hat is integrating OpenShell into its full-stack Red Hat AI platform to maintain infrastructure-level oversight and policy. The company is also making key contributions to the OpenShell upstream open-source project to help standardize the management of agents on enterprise platforms.

Yesterday’s announcements build on recent integrations by SAP, which is embedding OpenShell into Joule Studio runtime — part of SAP Business AI Platform for enterprise AI agents — and ServiceNow, which secured Project Arc, ServiceNow’s enterprise autonomous desktop agent, with OpenShell to add policy-based management for enterprise safety.

OpenShell runs in on-premises, hybrid, and enterprise cloud environments, local devices such as NVIDIA RTX Spark, NVIDIA DGX Spark, and GB10 systems from system providers, as well as NVIDIA DGX Station for Windows and NVIDIA DGX Station GB300 systems from NVIDIA partners.

• Huang’s keynote at NVIDIA GTC Taipei.

Anthropic has surpassed OpenAI as the world’s most valuable AI start-up after raising $65 billion, achieving a valuation of $900 billion, as the two San Francisco-based companies duel for supremacy in this technology. OpenAI, which kicked off the AI boom with the launch of ChatGPT in 2022, announced it had raised $122 villion in a funding round that put its value at $730 billion.

Both companies are planning to file for an initial public offering this year.

The new investment in Anthropic, led by Greenoaks Capital, Sequoia Capital, Altimeter Capital, and Dragoneer Investment Group, boosted Anthropic’s value two and a half times its previous valuation of $380 billion, about three months ago.

Anthropic has now raised more than $130 billion since its establishment. Its roster of investors includes Capital Group, Menlo Ventures, Iconiq Capital, and Lightspeed Venture Partners, as well as tech giants like Amazon and Google.

On the rise over the past few months, Anthropic is the creator of Claude, a chatbot (with Claude Opus 4.8 as its new flagship model), and Claude Code, code-writing software on which the company has built a robust business.

Recently, it released an AI security model, Mythos, capable of finding and exploiting hidden flaws in software. It also advised Pope Leo XIV on his encyclical warning about AI’s most disruptive effect.

By releasing new frontier models in April at higher API prices, Anthropic and OpenAI have found product-market fit by selling massive amounts of tokens across industries, according to an analyst. Coding agents burn a large number of tokens, and by charging $10-$20/month, the two start-ups won’t generate enough revenue to cover $1 trillion in infrastructure costs.

In a stark warning and global alarm, Pope Leo XIV called for action during an explosive speech at the Vatican yesterday.

Marking a rare break from papal tradition, the Pontifex unveiled his first encyclical, Magnifica Humanitas, at the Vatican, warning that artificial intelligence “needs to be disarmed to prevent domination, exclusion, and threats to humanity.”

Speaking at the Aula Nuova del Sinodo, the pope urged global cooperation to ensure AI serves peace, justice, and the common good. The landmark address compared the AI revolution to the industrial transformations the Church faced over a century ago.

The encyclical Magnifica Humanitas (or “Magnificent Humanity”), on the protection of the human person in the age of artificial intelligence, was signed by the Holy Father on May 15, 2026, the 135th anniversary of Pope Leo XIII’s encyclical Rerum Novarum (known in English as “Rights and Duties of Capital and Labor.”)

Here are some of Leo’s themes in the encyclical:

• AI is fundamentally not human.
We must avoid the misconception of equating this type of “intelligence” with that of human beings. These systems merely imitate certain functions of human intelligence. In doing so, they often surpass human intelligence in speed and computational capacity, offering tangible benefits across many fields. Yet this power remains entirely tied to data processing. So-called artificial intelligences do not undergo experiences, do not possess a body, do not feel joy or pain, do not mature through relationships, and do not know from within what love, work, friendship, or responsibility mean.

Leo describes the field of artificial intelligence as swiftly evolving, and with real promise as a “valuable tool.” But he emphasizes throughout the text that, on a profound level, artificial intelligence is not human, however closely it approximates the human mind and even its soul.

This view clearly differentiates between machines and humans. It directly counters a view of some A.I. researchers and thinkers, including some in the room who have recently raised questions about whether A.I. systems may actually feel or express human emotion.

• Humane labor practices and just wages remain essential.
The various kinds of job insecurity, fragmented career paths, and automation must not be evaluated solely in terms of efficiency, but in relation to the dignity of the worker, the right to sufficient remuneration, and the genuine possibility of participating in society.

AI has already displaced many entry-level jobs, and while the full scope of its impact is far from clear, the mass automation of both white-collar and blue-collar work is likely to significantly reshape most sectors of the labor market.

Echoing many of his predecessors, including Pope John Paul II, Leo acknowledges that economic and technological systems may undergo radical upheavals over the course of history, but insists that the essential dignity of the worker — which includes fair wages — must remain at the center of any new order.

In another section, he condemns “new forms of slavery” connected to the digital economy, including the young people who work for minimal pay in jobs like data labeling and content moderation, and the even younger ones who labor under dangerous conditions extracting the rare earth materials the industry requires: “The bodies of these people are scarred, injured and worn down so that computational flow may continue uninterruptedly.”

• No technology can take away the dignity of ordinary human beings.
We are living through a rapid phase of transition, a “change of era,” in which — while some are vying for the future of new technologies and others dedicate themselves to reflecting on the matter — most people are watching and waiting, observing from afar and merely hoping for the best.

The Vatican invited people from Silicon Valley to the formal introduction of the encyclical on Monday, including, notably, Christopher Olah, a co-founder of Anthropic, who participated in the presentation.

But the encyclical itself reminds readers that the aspiring history-makers in the room are not the only ones who have worth. Most of the world’s population will simply have to live with the fallout of how those leaders steward this technological revolution. “Magnifica Humanitas” insists that each of those people “observing from afar” matters.

“The value of persons, however, does not depend on what they achieve or produce,” Leo writes elsewhere in the text. “There are rights that apply to everyone simply by virtue of being human.” The document uses the word “dignity” 100 times.

• Beware the temptation of erecting a new Tower of Babel.
With the heart of a shepherd and a father, I ask everyone to abandon the construction of yet another Tower of Babel and to join forces in building up the common good, so that humanity will never lose its beauty, and the world once again will come to recognize the human heart as the place where God desires to dwell.

The biblical story of the Tower of Babel recurs as a touchstone. The account appears in the Book of Genesis and describes a world in which a unified human population that speaks only one language decides to build a tower “whose top reaches to the heavens” to exert its own power and domination.

In response, God scatters the people across the earth, in what serves as an origin story for the existence of different languages and cultures.

Leo uses the Tower of Babel as an illustration of the pitfalls of pursuing uniformity and standardization, and the limits of ambitious undertakings that appear able to compete with the claims of religion. As many aspects of global culture homogenize, and technology becomes a kind of universal language, Leo’s call for humility and diversity stands in contrast. It’s also a reminder that many of the seemingly new ethical and social challenges posed by A.I. have ancient roots.

• The pope cites research and makes concrete recommendations.
In recent years, psychological and psychiatric literature has documented with growing insistence how early and unsupervised exposure to digital devices and social media can negatively impact sleep, attention span, control of emotions and relationships, especially during the most vulnerable stages of life, at times with tragic consequences.

For all its sweeping moral force, “Magnifica Humanitas” is also a practical document, showing how Leo is focused on pastoral care for the church’s hundreds of millions of families. It surveys research on the impact of technology on child development, including how early and unsupervised access to cellphones leaves children vulnerable to addiction, bullying, and sexual exploitation. Other topics include the regulation of data ownership and the use of A.I.-related weapons in war.

• Human life is beautiful.
For this reason, humanity — in all its grandeur and woundedness — must never be replaced or surpassed. We can embrace technological progress that alleviates suffering and unlocks new possibilities, provided we do not abandon the very essence of our humanity, namely, the capacity for relationship and love.

The title of “Magnifica Humanitas” says it all: In the end, Leo is less interested in technology than in humanity. Humans are flawed, vulnerable, and finite, the pope writes. We are increasingly inferior to the technology we have created if we measure only in cold terms of performance. But the pope writes with great affection for humans. The text ends with a wish “that we may bear witness to the grandeur of humanity, in which God has made his dwelling.”

Humanity, created by God in all its grandeur, is today facing a pivotal choice: either to construct a new Tower of Babel or to build the city in which God and humanity dwell together. In Jesus Christ, this humanity in its grandeur becomes the Way, the Truth and the Life, opening…

— Pope Leo XIV (@Pontifex) May 25, 2026

Using the Gemini 3.5 Flash model, Google is increasing its search box size and making it more interactive so that people can ask even longer questions, upload photographs and videos into queries, and ask follow-up questions.

It’s the first overhaul of its iconic search bar in 25 years, since 2001, prompted by the rise of AI.

In addition to adding a chatbot on the main search page, Google will also offer digital assistants, or agents, to automate searches.

For example, someone apartment hunting can be notified of a new listing without opening the real estate website Zillow.com.

Google’s CEO, Sundar Pichai, said at the company’s annual developer conference in Mountain View, Calif., this month, “When people use our AI in search, they use search more.”

This move shows that Google has solidified its position as an AI heavyweight.

In addition to its Gemini models, it is producing AI chips and pouring hundreds of billions of dollars into data centers for its cloud computing business.

Its Gemini app, which can do coding and research, now has 900 million active users — about the same number as ChatGPT.

Analysts say that these changes are helping Google make more money from advertising. Last year, Google’s ad clicks rose 6 percent, and it charged 7 percent more for each click. The company’s annual profit has more than doubled since 2022 to $132 billion.

On searches that deliver AI Overviews, people can ask follow-up questions in AI Mode, which Mr. Pichai called “a revelation.”

Google is also bringing one of AI’s biggest breakthroughs — software coding — to search.

Google said it was introducing an alternative to the agents powered by Anthropic’s Claude Code and OpenAI’s Codex.

Called Gemini Spark, the service is embedded in Gmail, Docs, and other Google products, where it can turn meeting notes spread across emails and chats into a single document. It can also read and draft emails.

Google’s AI-driven shopping cart will also recommend discounts when products go on sale and warn people when they select items that could be incompatible.

Koray Kavukcuoglu, the chief technology officer at Google DeepMind, the company’s AI lab, said that “plugging Gemini into Google’s products will help the company stay ahead of competitors by providing information about users’ needs.”

• Google Issues “Gemini Omni “, a New Model that Can Create Anything from any Input

Anthropic announced this month the launch of over 20 new MCP connectors and 12 plugins tailored to specific legal work and practice areas, for paying attorneys, law students, and others in the legal sector.

These tools link Claude to dedicated legal software to tackle specific workflows. They are intended for handling day-to-day legal document work, including contract review, drafting, redlining, extraction, and comparison. o

One feature, called “Commercial Counsel”, reviews vendor agreements, while another tool is intended to help with studying for the bar exam.

This setup connects to third-party software services commonly used in law, such as DocuSign, Thomson Reuters, and even a competing AI legal service like Harvey.

Each agent template can be installed in Cowork or Claude Code with a click and produces outputs that match institutional drafting standards.

The plugin and skill ecosystems are open protocols, and early contributors, including Box, Legal Quants, Lawve AI, and Thomson Reuters, have already shipped their own skills, plugins, and style conventions. Any partner can submit connectors and skills through the Directory. ‍

Anthropic is also competing with OpenAI in other industries, such as financial services and health care.

Both AI companies are trying to convince potential customers of their offering on security, data privacy, and questions related to IT teams.

AI frontier models are seeking to move beyond infrastructure, software licensing business, and API services in an AI-first world.

They are launching private equity firms backed by institutional capital and are willing to enter the enterprise workflows business by moving closer to implementation, orchestration, and transformation work — an area that has long been dominated by IT outsourcing firms.

These AI-native companies increasingly want to own more of the enterprise workflow layer, while enterprises themselves want greater control over strategic technology capabilities, specifically execution, governance, and accountability.

On May 4, Anthropic unveiled a $1.5 billion venture backed by investors, including Blackstone, Goldman Sachs, Hellman & Friedman, and Sequoia Capital.

On the same day, it was disclosed that OpenAI was raising over $4 billion for its own initiative, “The Development Company,” at a reported valuation of $10 billion.

Weeks earlier, Google Cloud announced strategic partnerships with Vista Equity Partners and CVC. It is also reportedly exploring arrangements with Blackstone, KKR, and EQT.

Analysts say this approach increasingly resembles the “forward-deployed engineer” model popularised by Palantir, where software companies move beyond selling technology and embed themselves deeply within enterprise operations.

The model is based on deploying forward-deployed engineers — a kind of AI deployment engines — to work side by side with portfolio companies of private equity firms, helping them build and optimize AI solutions on top of its models and broader AI stack.

Karthik Narain, Chief Product and Business Officer at Google Cloud, said that these partnerships would accelerate AI adoption across sectors and drive industry-wide digital transformation. Anthropic is targeting mid-sized companies in sectors like healthcare, manufacturing, financial services, retail, real estate, and infrastructure.

“Demand for hands-on AI implementation in sectors is significantly outpacing what is available across the industry today,” he said.

Krishna Rao, chief financial officer at Anthropic, said this month that the company’s partnerships remain central to how Claude reaches large enterprises, and that it continues to “invest deeply” in those relationships.

Currently, enterprises need engineers to migrate infrastructure, modernize applications, integrate systems, and manage increasingly complex digital estates. At the same time, implementation work, coding, testing, support, maintenance, and orchestration are increasingly becoming automatable.

Boosting AI adoption among enterprises is a strategic necessity for Anthropic and OpenAI as they seek to bolster revenue growth and justify their sky-high valuations ahead of their highly anticipated initial public offerings, which can come this year.

For Google, the move is equally crucial to maintain the rapid growth momentum at Google Cloud, which is emerging as a key growth engine for the tech giant’s revenues and profits.

During the company’s earnings call this month, Alphabet CEO Sundar Pichai said that enterprise AI solutions became the primary growth driver for Google Cloud for the first time in the past quarter.

At least 18 federal class-action lawsuits have been filed against Canvas LMS’s parent company, Utah-based Instructure, following last week’s data breach, while the company issued an apology after final exams at numerous universities were halted.

“Over the past few days, many of you dealt with real disruption. Stress on your teams. Missed moments in the classroom. Questions you couldn’t get answered. You deserved more consistent communication from us, and we didn’t deliver it. I’m sorry for that,” Instructure CEO Steve Daly wrote in a new post.

The hacking group behind the outage, ShinyHunters, also stole data on potentially tens of millions of students across nearly 9,000 schools.

The company’s ongoing investigation found that “usernames, email addresses, course names, enrollment information, and messages” were exposed. That’s slightly different from Instructure’s initial findings, which said that “names, email addresses, student ID numbers” had been affected. That said, usernames and email addresses can still expose a student’s full name.

Daly added: “We’re still validating all findings, but we want to be clear about what we understand was and wasn’t affected.”

Instructure’s CEO also revealed that hackers exploited a “vulnerability regarding support tickets in our Free for Teacher environment,” a service that enables teachers to use some Canvas services at no cost.

In response to the hack, Instructure has temporarily shut down the Free-for-Teacher service.

• The great 2026 Canvas-ocalypse by Bryan Alexander

• Instructure Is Risking the Trust That Built Canvas by Phil Hill

The AI Agent Conference 2026, curated by Firsthand VC in partnership with NYSE Wired, Bright Data, and theCUBE, drew over a thousand senior executives, AI engineers, and investors to Midtown Manhattan, May 4–5, 2026, for two days, who effectively declared the experimentation phase of agentic AI over.

The message from the stage and the hallway alike was the same: the conversation had moved from “should we deploy agents?” to “how do we govern, secure, and scale them without getting fired?”

However, one of the most telling figures of the conference was this: while 79% of organizations report some level of agent adoption, only 11% are running agents in production.

Another cautionary figure indicated that 40% of projects are at risk of cancellation, and only 6% of organizations qualify as true AI high performers. Closing that gap was the conference’s defining theme.

Organized around three interconnected tracks — Agentic Enterprises, Agentic Engineering, and Agentic Industries — plus a specialized Web Discovery & Execution track, the conference landed at the exact inflection point where enterprise adoption is crossing from pilot programs into operational commitment.

Alongside it, The Agentic List 2026 was unveiled: a curated ranking of 120 companies across three themes (Enterprises, Engineering, Industries), selected from over 5,000 nominations across nearly 2,000 screened companies, collectively backed by billions in funding.

Notable inclusions were:

Agentic Enterprises: Glean ($765M raised), Perplexity ($976M), Ramp ($2.8B), Apollo ($251M), Clay ($202M), Sierra ($635M), Decagon ($481M)

Agentic Engineering: Mistral AI ($3.2B), Cohere ($1.5B), n8n ($254M), Cognition ($596M), Augment Code ($252M), CrewAI ($18M), LangChain ($160M), Tavily ($25M, recently acquired by Nebius)

Agentic Industries: AlphaSense ($1.4B), Hippocratic AI (healthcare), Hebbia (legal), Harvey (legal)

The Numbers That Framed Sessions

The conference leaned hard on data, and speakers referenced these staggering figures throughout both days:

• $10.8–12 billion: Projected global agentic AI market size in 2026, growing at a 43.8–46% CAGR toward $139–196 billion by 2034 (Grand View Research, Precedence Research, Allied Market Research)
• $301 billion: Total global AI spending in 2026, with agentic AI representing 10–15% of enterprise IT budgets (IDC)
• 40%: Enterprise applications that will embed task-specific AI agents by year’s end — up from under 5% in 2024 (Gartner)
• 171%: Average ROI from enterprise agentic AI deployments globally; 192% for U.S. enterprises specifically (Deloitte 2026 State of AI in the Enterprise)
• 5.8x: Average ROI on AI investment within 14 months of production deployment (McKinsey)
• $4.6 million: Average annual savings per enterprise from AI-driven process automation across 3+ departments (McKinsey / IDC)
• 88%: Organizations now using AI in at least one function, up from 78% the prior year (McKinsey / Gartner)
• 100%: Of surveyed enterprises planning to expand agentic AI usage in 2026 (CrewAI survey of 500 C-level executives at $100M+ revenue organizations).”
• 6%: Organizations that qualify as true AI high performers with more than 5% of EBIT attributable to AI (McKinsey)

Day 1: Vision Meets Infrastructure

Opening keynotes set the tone with two back-to-back sessions spanning the full arc of the enterprise agent story.

Ameet Talwalkar, Chief Scientist at Datadog, kicked off the event, presenting how Datadog is rebuilding its observability platform to treat agents as first-class citizens alongside human users and traditional applications.

Arvind Jain, Founder and CEO of Glean — which surpassed $200 million in ARR at a $7.2 billion valuation after its $150 million Series F — took the stage alongside Sapphire Ventures’ Jai Das, defending that enterprise agents need a unified context layer connecting LLMs to internal business data, and saying that this approach has become the blueprint for how many large organizations when dealing with an agentic deployment.

“The real bottleneck is not the models themselves but connecting their reasoning power to the context inside your company,” he said.

Joe Moura, Co-Founder and CEO of CrewAI, said, “Enterprise adoption of agentic AI is accelerating faster than anyone anticipated. Organizations aren’t just experimenting — they’re building, shipping, and scaling agents into production.”

UiPath’s Raghu Malpani, Chief Product and Technology Officer, stated: “We’re at a pivotal moment where AI, deterministic automation, and orchestration are coming together to reshape how work gets done.”

Day 2: No Protocol War

During the Agentic Engineering track, a consensus emerged around two interoperability protocols that are rapidly becoming the backbone of enterprise agent infrastructure.

1. MCP (Model Context Protocol), created by Anthropic in November 2024 and donated to the Linux Foundation’s Agentic AI Foundation, has become the standard interface connecting agents to external tools, databases, and APIs. By early 2026, MCP had crossed 97 million monthly SDK downloads, with adoption from every major AI provider, including OpenAI, Google, Microsoft, and Amazon. Its architecture is client-server via JSON-RPC 2.0.
2. A2A (Agent2Agent Protocol), launched by Google in April 2025 and now governed by the same Linux Foundation body, handles the other half — how agents communicate with each other across organizational and platform boundaries. Google announced at Cloud Next 2026 that A2A has reached version 1.2 and is running in production at 150+ organizations, including Microsoft, AWS, Salesforce, SAP, and ServiceNow. Its architecture is peer-to-peer via HTTP and Server-Sent Events.

Multiple panelists emphasized that these are not competing standards. MCP handles vertical integration (agent-to-tool), while A2A handles horizontal coordination (agent-to-agent). The dominant question was how fast developers can implement both.

Google’s broader moves were discussed. The rebranding of Vertex AI to the Gemini Enterprise Agent Platform, the launch of Workspace Studio as a no-code agent builder, and the introduction of Agentic Data Cloud collectively represent Google’s bid to own the full stack from chip to inbox.

Salesforce’s Agentforce loomed over many conversations. It has reached $540 million in ARR with 18,500 enterprise customers. Speakers shared that Agentforce already autonomously resolves 70% of customer chats for clients like 1-800Accountant during peak seasons.

Dominant Theme of Security and Governance

CrewAI’s survey found that security and governance ranked as the #1 priority (34%) when enterprises evaluate agentic AI platforms. Ease of integration came second at 30%, and reliability came third at 24%.

Enterprises question whether agents can be deployed safely at scale without exposing unacceptable risk. Data backed this concern:

• 40%+ of agentic AI projects are at risk of cancellation by 2027 due to governance and ROI gaps (Gartner)
• 25% of enterprise breaches by 2028 will be traced to AI agent abuse (Gartner)
• 76% of enterprises cite data privacy and security as their top AI risk concern (IDC)
• 68% cite lack of identity security controls for AI specifically (IDC)
• Only 21% of organizations have a mature governance model for autonomous AI agents (Deloitte)
• 64% of CEOs acknowledge that FOMO drives AI investment before fully understanding the value (IBM)
• $2.1 billion in regulatory fines related to AI misuse were issued globally in 2025 — a 7x increase from 2023

CrowdStrike’s Atul Tulshibagwale and AgentCloak’s Peter Yared presented on agentic identity and the real attack surface of connecting agents to thousands of external MCP servers — including tool-poisoning attacks and data-exfiltration risks.

The concept of “Guardian Agents” — autonomous systems whose sole purpose is to monitor, oversee, and constrain the behavior of other agents — drew both enthusiasm and skepticism. Gartner projects 40% of CIOs will demand Guardian Agents by 2028.

Finance and Healthcare Lead Industry Deployments

• Steve Hasker, President & CEO, Thomson Reuters, demonstrated that many legacy information giants are betting their futures on autonomous workflows, not incremental copilot features.

• Rob Wisniewski, CTO of Credit & Insurance, Blackstone, discussed how agentic AI is moving beyond back-office automation into core deal-making and underwriting.

• Sirisha Kadamalakalva, MD, Global Head of AI/ML Investment Banking at Citi, covered agent deployment in regulated financial workflows.

• Karun Appapogu, Head of AI Technologies Architecture, Vanguard, and Kevin Hearn, SVP, Axos Bank, added retail and digital banking perspectives.

• Shipali Jangra, Director of Global Digital Product Management at American Express, spoke to operational scale.

Over 78% of financial services organizations have adopted AI agents.

• Commerzbank’s Microsoft-powered banking assistant resolves 75% of customer requests across 30,000+ monthly conversations.

• Allianz Partners is targeting 90% autonomous operations across claims and invoice processes spanning 1,000+ FTEs.

Healthcare showed equally compelling results.

• AtlantiCare‘s agentic AI clinical assistant achieved an 80% adoption rate among test providers, cut documentation time by 42%, and freed approximately 66 minutes per clinician per day.

• AI-powered imaging solutions are expected to prevent up to 2.5 million diagnostic errors annually. At the enterprise level, around 65% of healthcare organizations have adopted AI agents.

Commerce had its own dedicated sessions featuring Felipe Romano (PayPal), Robin Chiang (OpenTable), and Richard Cohene (Lightspeed Commerce).

In manufacturing, Samsung has committed to transforming all its facilities into AI-driven factories by 2030. Fujitsu’s AI development platform, launched in early 2026, reduces software modification time from three months to four hours, according to the company.

Seven Trends That Defined the Conference

1. Multi-Agent Orchestration Goes Mainstream
66.4% of the agentic AI market now focuses on coordinated multi-agent systems rather than single-agent solutions. LangGraph, CrewAI, and Google’s ADK are converging around graph-based and role-based orchestration models. Multi-agent systems are projected to grow at a 48.5% through 2030.

2. Context Engineering Replaces Prompt Engineering
Salesforce and others emphasized that agent performance depends less on how you ask a question and more on the information architecture surrounding the agent — which data sources it can see, how context is structured, and what gets retrieved when.

3. Agent Identity Becomes a Product Category
CrowdStrike, AgentCloak, Descope, C1, and others are building standalone products for authentication, authorization, and audit trails specifically for autonomous agents.

4. Headless AI and API-First Architectures
Salesforce’s Headless 360 and Google’s agent platform signal a shift from traditional UI-driven software to API-first architectures where agents access platforms programmatically rather than through dashboards. By 2028, one-third of user experiences will shift from native apps to agentic front ends.

5. The Agent Framework Landscape Consolidates
Framework adoption nearly doubled, rising from 9% to 18% of organizations (Datadog). LangGraph, CrewAI, and OpenAI’s Agents SDK are emerging as the dominant choices.

6. The Workforce Transformation Is Real
McKinsey estimates 44% of U.S. work could be performed by AI agents with current capabilities. 66% of enterprises are reducing entry-level hiring as they deploy AI. 77% of employers plan to upskill workers. AI/ML engineers’ salaries reached a median of $185K in the U.S., with demand up 74%.

7. The Governance Gap Is Existential
Over 40% of agentic AI projects risk cancellation by 2027. Only 21% of organizations have mature AI governance. $2.1B in AI-related regulatory fines were issued in 2025 alone. The EU AI Act is already forcing 42% of global enterprises to adjust practices.

What Enterprise Should Do

Several sessions converged on a practical playbook:

1. Audit your current agent deployments — move beyond pilot metrics to production-grade KPIs, including completed tasks, escalation rates, and resolution speed
2. Implement both MCP and A2A protocols to future-proof agent infrastructure for cross-vendor interoperability
3. Establish an agent governance framework now — define identity, scope, audit trails, and escalation thresholds before scaling
4. Prioritize context engineering over prompt engineering — invest in retrieval quality, data architecture, and information design
5. Start where ROI is clearest: customer service, eCommerce, finance automation, and software engineering are the proven winners
6. Budget for agentic AI security — tool poisoning, prompt injection, and data exfiltration are real attack surfaces
7. Fix data foundations first — 52% of organizations cite data quality as their primary blocker, and IDC predicts a 15% productivity loss by 2027 for companies that fail to establish AI-ready data foundations

“Data quality, not model size, is the primary bottleneck in AI performance,” said Datology’s CEO, Ari Morcos, at the ODSC AI East 2026 in Boston this week. “Better training data and smaller models outperform larger ones trained on slop,” he explained.

Ami Bhatt, FDA Chief Innovation Officer and Chair of the American College of Cardiology, discussed AI in clinical decision support and the FDA’s evolving framework for validating AI systems in healthcare. “At the FDA, we are building regulatory infrastructure for AI — not blocking it, but demanding rigor.”

Across healthcare, finance, government, legal, education, manufacturing, and energy, a pattern emerged: the organizations moving fastest on AI are the ones that have solved the data privacy and deployment ownership equation, performing at frontier quality while keeping patient data locked down. They’re not asking “which model” but “where does it run, who owns the data, and can we audit every decision?”

By industry sector, the ODSC East 2026 generated several outcomes:

In Healthcare and Biopharma, AI in drug discovery dominated. Generative AI for molecular design is now producing novel candidate compounds in hours instead of months. Multiple sessions covered AI-driven biomarker discovery — using foundation models to identify disease signatures in genomic data that traditional bioinformatics pipelines miss entirely.

At major health systems, medical imaging AI has evolved into predictive modeling for clinical outcomes.

Healthcare sessions circled back to the issue of data privacy, as foundation models require massive datasets to perform, but HIPAA, patient consent, and institutional data governance impose hard constraints on data sharing. Synthetic data generation and federated learning emerged as the most discussed workarounds, but neither is mature enough for enterprise-scale deployment yet.

In Government and Defense, federal agencies are moving from “should we use AI?” to “how do we deploy AI in air-gapped, classified, and compliance-heavy environments?” FedRAMP, NIST 800-53, and ITAR requirements dominated the conversation.

The agencies that are moving fastest — DOD, intelligence community, DHS — are the ones with the most acute operational pain points and the budget to solve them.

Google’s signing a classified AI deal with the Pentagon (reported during the conference week) added urgency to the discussion. The open question was whether the government AI infrastructure should be owned by hyperscalers or if agencies would build sovereign capability.

An important takeaway was that only American-made models and compliance-first would have a massive moat.

In Financial Services, a recurring theme was zero tolerance for AI hallucinations, as they could be a compliance violation. Multiple sessions addressed guardrails, output validation, and human-in-the-loop architectures specifically designed for regulated environments.

Another takeaway is that the shift to open source in finance was real, and self-hosted, private deployment is becoming the default architecture. Several financial services practitioners described their institutions moving from proprietary APIs (OpenAI, Anthropic) to self-hosted open-weight models (Llama, DeepSeek, Mistral), driven not by cost but by data sovereignty.

When your trading strategies, M&A documents, and client portfolios are the data, sending them to a third-party API is a non-starter.

Financial services want frontier-quality AI but absolutely cannot accept the data exposure inherent in shared platforms.

In Manufacturing and Supply Chain, AI for robotics reflected the convergence of foundation models with physical systems. Sessions covered reinforcement learning for warehouse automation, computer vision for quality inspection, and multi-agent coordination for logistics.

The manufacturing story is about integration, not intelligence. The models are capable enough — the bottleneck is connecting AI to legacy ERP systems, SCADA networks, and supply chain databases that were built decades before APIs existed. Several sessions addressed the “last mile” problem of getting AI outputs into SAP, Oracle, and custom MES systems.

Participants agreed that manufacturing AI is a data integration challenge first and a model challenge second. The organizations winning here are the ones who’ve invested in data infrastructure, not just model training.

In Legal and Compliance, the recurring pattern was that law firms and in-house legal teams were deploying LLMs for document review, contract analysis, and legal research, but with extreme caution.

Attorney-client privilege is the hard constraint. Unlike other industries where data privacy is a regulatory concern, in legal it’s a constitutional one. Multiple speakers from regulated industries described building air-gapped AI systems specifically so privileged communications never touch external infrastructure. The phrase “private deployment” came up more in legal-adjacent sessions than anywhere else.

AI governance frameworks — as Shoshana Rosenberg presented — are being adopted fastest by legal departments, not IT departments. Lawyers understand regulatory risk intuitively and are building the policies and controls that other functions are still debating.

Regarding Energy, Utilities, and Infrastructure companies, practitioners highlighted the need for AI that can be deployed reliably and deterministically on local hardware, without cloud dependencies, given that operations often occur on factory floors, in substations, and on drilling platforms, where connectivity is unreliable or prohibited.

In terms of Software Engineering and Devtools, developers at the conference analyzed how to evaluate whether AI-written code is correct, secure, and maintainable. Evaluation systems, test generation, and multi-agent code review (in which multiple AI agents check each other’s work) were the most-discussed engineering patterns.

Karen Zhou from Anthropic’s Claude Code team and Robert Brennan from All Hands AI (OpenHands/OpenDevin) enlightened the discussion.

Software developers mostly agreed that competitive edge was in AI systems that can reason across codebases, call external services, and operate production infrastructure — not just autocomplete.

Finally, in Higher Education and Research, Brown University’s Michael Littman, Boston University’s Mohammad Soltanieh-ha, Bentley University’s Noah Giansiracusa, and MIT’s Max Tegmark delivered major sessions. The academic-to-production pipeline has never been shorter.

Many institutions are now building their own AI operating systems — course-specific agents, research assistants, administrative automation tools, and student support chatbots. Multiple sessions referenced institutions deploying self-hosted LLMs on their own cloud infrastructure to maintain FERPA compliance and student data privacy.

One takeaway was that institutions that own their AI infrastructure, rather than subscribing to shared platforms, would emerge as the leaders.