🔐 Anthropic Warns Government: Mythos Makes Cyberattacks 'Much More Likely'
ai-security
Anthropic is privately briefing top government officials that its unreleased Mythos model is "far ahead of any other AI model in cyber capabilities" — and that large-scale cyberattacks are now much more likely because of it. GPT-5.3-Codex was already classified "high capability" for cyberattacks under OpenAI's Preparedness Framework in February, but this is the first time a frontier lab has proactively warned policymakers that its own unreleased model creates an elevated national threat.
🏗️ Amazon + OpenAI Co-Create Stateful Runtime on Bedrock
llm-infrastructure
Amazon and OpenAI announced a joint Stateful Runtime Environment on Amazon Bedrock — persistent agent memory, tool call state, and multi-step execution across sessions as first-class infrastructure. This isn't a model launch; it's AWS declaring that runtime and memory are the product category, and partnering with the model maker to own the layer beneath the intelligence.
💰 $3.6B in Agentic AI Security Funding, $96B in M&A at RSAC 2026
ai-security
Palo Alto, ServiceNow, and Alphabet spent a combined $72B on cybersecurity M&A in 12 months; agentic AI security startups raised $3.6B; the total disclosed MCP security funding — for the protocol connecting agents to enterprise systems — sits at just $40M. The gap between where incumbents are buying and where the market is funding tells you exactly where enterprise security is heading in 2026.
📡 More signal, less noise → www.thesignal.press