AI News Roundup: Mythos Reshapes Cybersecurity, Pentagon Signs 8 AI Deals, Connecticut Passes Landmark AI Law

Anthropic’s Mythos Is Rewriting the Cybersecurity Playbook

The biggest story in AI right now isn’t a chatbot — it’s a vulnerability hunter. Anthropic’s Claude Mythos, first revealed in late March, has identified thousands of previously unknown zero-day vulnerabilities across every major operating system and web browser. The model can find, reverse-engineer, and exploit software flaws at industrial speed, including in closed-source applications.

Under Project Glasswing, Anthropic has limited access to a handful of American companies — Apple, Amazon, JPMorgan Chase, and Palo Alto Networks among them — plus over 40 organizations that maintain critical infrastructure software. The goal is purely defensive: scan and secure systems before bad actors can exploit the same class of vulnerabilities. Anthropic CEO Dario Amodei has publicly warned of a “moment of danger” as the gap between AI offensive capability and defensive readiness widens.

OpenAI responded within weeks by unveiling GPT-5.5-Cyber, a cybersecurity-specific model with limited access for vetted security teams. The EU has requested access to both models, but Anthropic is holding firm on its restricted release strategy.

Pentagon Signs AI Deals with 8 Tech Giants — Anthropic Excluded

The U.S. Department of Defense announced agreements with SpaceX, OpenAI, Google, NVIDIA, Reflection AI, Microsoft, Oracle, and Amazon Web Services to deploy AI on classified Impact Level 6 and 7 networks. More than 1.3 million DoD personnel already use GenAI.mil, the Pentagon’s secure enterprise AI platform, and these deals extend large language models into classified operational workflows for warfighter decision-making and situational awareness.

Notably absent: Anthropic. The AI lab insisted on guardrails preventing its technology from being used for mass surveillance and autonomous weapons systems. The Pentagon wanted unrestricted operational use. The impasse led the DoD to diversify its vendor base — a decision that now looks prescient given Mythos’s capabilities, but one that leaves the department without access to what many consider the most advanced AI system in the world.

Connecticut Enacts One of America’s Most Comprehensive AI Laws

Connecticut’s bipartisan SB 5 cleared the state legislature with overwhelming margins — 131–17 in the House, 32–4 in the Senate — and Governor Ned Lamont is expected to sign it imminently. The omnibus bill addresses automated employment decisions, AI companion safety for minors, synthetic media transparency, and whistleblower protections for frontier model developers.

Starting October 1, 2026, companies using AI as a “substantial factor” in hiring, promotion, or termination must notify affected workers and disclose the technology’s purpose, data categories, and sources. The law also requires operators of AI companion chatbots to implement industry-leading safeguards against harmful interactions with minors — including romantic or sexual content, encouragement of self-harm, and manipulative techniques designed to foster emotional dependence. A temporary cure period runs through December 2027, enforceable exclusively by the state attorney general.

Trump Administration Reverses Course on AI Oversight

In a significant policy pivot, the Trump Administration is now actively considering oversight mechanisms for advanced AI models — a reversal from its earlier anti-regulation stance. The shift was largely triggered by the national security implications of Anthropic’s Mythos, which demonstrated that frontier AI can identify and exploit critical infrastructure vulnerabilities at a scale and speed that existing defenses cannot match.

In parallel, Google, Microsoft, and xAI have agreed to give the U.S. government early pre-release access to their AI models through the Commerce Department’s Center for AI Standards and Innovation, joining OpenAI and Anthropic. The administration’s March 2026 National Policy Framework called on Congress to establish a unified federal approach that would override the growing patchwork of state-level AI laws — a position that gains urgency as states like Connecticut move ahead on their own.

OpenAI Expands ChatGPT Ads to Global Self-Serve Platform

OpenAI opened ChatGPT Ads Manager to self-serve advertisers this month, eliminating the previous $50,000 minimum spend and introducing cost-per-click bidding with expanded measurement tools. Ads appear in labeled, subtly tinted boxes below AI responses and explicitly do not influence the content of those responses.

The geographic footprint is expanding rapidly: the U.K., Brazil, Japan, South Korea, and Mexico join the U.S., Canada, Australia, and New Zealand as active markets. OpenAI has set an ambitious target of $2.5 billion in ad revenue for 2026 and $100 billion by 2030 — figures that would make ChatGPT one of the largest advertising platforms in the world within four years.

Isomorphic Labs Raises $2.1B to Scale AI Drug Discovery

DeepMind spinoff Isomorphic Labs closed a massive $2.1 billion Series B round, underscoring investor confidence in AI-driven pharmaceutical research. The company, founded by Demis Hassabis, uses AI models descended from AlphaFold to predict molecular interactions and accelerate drug candidate identification. The round was backed by Thrive Capital, GV, Temasek, and HarbourVest.

The funding comes as the broader AI-pharma sector heats up. Novo Nordisk’s recent partnership with OpenAI, combined with existing deals at Sanofi, Moderna, and Eli Lilly, suggests the pharmaceutical industry is moving decisively from AI experimentation to production-scale deployment.