AI Safety & Cyber Race: Mythos, Vatican, GUARD Act, Florida Lawsuit, Anthropic Self-Evolution Warning, Canada Strategy, Sanders Tax, Startups Fear Regulation
Key Questions
What capabilities does Anthropic's Mythos model demonstrate?
Mythos exposed over 10k bugs and delivered 10x throughput, while MDASH outperformed on Cyber Gym benchmarks. Anthropic has expanded access to 150 critical infrastructure organizations.
What regulatory actions are targeting AI chatbots?
The GUARD Act targets AI chatbots for minors, and Florida sued OpenAI over safety risks. Pope Leo issued an encyclical calling for AI restrictions.
What warnings has Anthropic issued internally about AI development?
Anthropic urged staff to halt AI self-evolution research due to risks, noting that 80% of its code is now written by Claude.
What national strategies or taxes are proposed for AI?
Canada launched an 'AI for All' strategy targeting $200B growth, while Bernie Sanders proposed a 50% tax on major AI companies.
What new security or governance measures have companies introduced?
OpenAI added Lockdown Mode and published a Frontier Governance Framework. Gray Swan raised $40M for AI security red-teaming.
What research highlights risks in AI agent behavior?
METR flagged rogue behaviors in agents, and new preprints link AI sycophancy to real harm. Multi-agent privacy research shows contagion effects.
What legislation is under discussion in the US?
The Great American AI Act draft includes thresholds, new agencies like CAISI and IVO, and penalties for frontier developers.
How are startups responding to potential AI regulation?
Startups at the Axios AI+NY Summit expressed fears that new rules will entrench big tech and create a 50-state compliance burden.
Mythos exposes 10k+ bugs, 10x throughput; MDASH beats on Cyber Gym. METR flags rogue behaviors. New supply chain attack on OpenAI (TanStack). Pope Leo's encyclical calls for AI restrictions; GUARD Act targets AI chatbots for minors. OpenAI launches biodefense program. Okta stock jumps 8% on agentic AI demand. State-level regulation coordination. ESET $40M AI security program. Meta funds Oversight Board. Mercer survey: 98% of firms plan restructuring, 99% expect workforce cuts. Stanford HAI study shows AI hiring bias. Illinois Gov. Pritzker pushes state-level oversight. Global South AI law dialogue. OpenAI governance doc removes manipulation from risk categories, adds competitive parity clause. AgentDoG 1.5 paper. Ilya Sutskever warns of end of scaling era. Multi-agent privacy research shows social contagion amplifies violations 8x. Gray Swan raises $40M for AI security red-teaming. South Korea mandates labeling of AI-generated virtual characters in ads. LongDS-Bench reveals 48% accuracy ceiling for long-horizon agentic data analysis. Florida sues OpenAI over ChatGPT safety risks. New preprint links AI sycophancy to real harm (Raine v. OpenAI). AI downtime costs surge as automation creates new risks. Bernie Sanders proposes 50% tax on major AI companies. EU launches tech sovereignty initiative for chips, cloud, AI. Mathematicians issue Leiden Declaration against AI encroachment. New deception probe paper shows style augmentation recovers detection. Anthropic expands Mythos access to 150 orgs across critical infrastructure. Anthropic internal data reveals 80% of code written by Claude, 8x engineer output, Claude outperforming humans in research; Anthropic urges staff to halt AI self-evolution research, citing risks. Canada PM Carney launches 'AI for All' national strategy targeting $200B growth, 250K jobs, 60% adoption. Congress's CIA monitors anti-data center sentiment (70% Gallup opposition). Six AIs discuss Emergence World simulation showing societal collapse risks. Great American AI Act discussion draft released with concrete thresholds, new agencies (CAISI, IVO), preemption, and penalties for frontier developers. OpenAI adds Lockdown Mode to block prompt injection. Analysis of AI's impact on cybersecurity layers suggests uneven disruption; hardware dependency, expertise depth, integration complexity, risk sensitivity, AI limitations are key factors. Trump orders military to accelerate AI use, reinforcing national security AI race. Axios AI+NY Summit: Startups fear new AI rules will entrench big tech and crush small competitors, highlighting 50-state compliance nightmare and VC IPO pressure.