Pentagon designates Anthropic a supply-chain national security risk after refusing autonomous weapons and surveillance access; Sam Altman publicly defends his rival; Perplexity launches Computer, a 19-model super agent; Anthropic publishes the AI Fluency Index showing 85% of users iterate on responses but scrutiny drops for polished outputs; and just 4% of companies have achieved ROI on AI investments.

Anthropic rejects the Pentagon's final offer on AI safeguards, refusing to remove guardrails for surveillance and autonomous weapons; Goldman Sachs says AI investment added basically zero to US economic growth in 2025; Google launches Nano Banana 2 image model with Pro-quality results at Flash speed; Ladybird browser uses AI agents to port 25,000 lines of C++ to Rust in two weeks; and the agent economy raises an unanswered question — when AI signs an SLA and doesn't deliver, who settles it?

The FDA clears the first human age-reversal trial — Sinclair's epigenetic reprogramming study launches next month; a rhetoric professor gives AI slop a formal seven-layer academic definition; a new paper coins UNDCS to name the regulatory gap agentic clinical AI is falling through; ICU AI decision support is outpacing nursing training and liability frameworks; and Auburn engineers watch WarGames to understand AI risk.

AI-designed drug candidates are hitting 80-90% Phase I success rates versus a 50% historical average; poet Sasha Stiles' AI collaboration "A Living Poem" lands at MoMA; CoreWeave and Nebius flagged as potential doubles for 2026; Sam Altman compares AI training energy to raising a human child; and Demis Hassabis proposes the Einstein Test as a genuine benchmark for AGI.

Anthropic publicly accuses DeepSeek, Moonshot AI, and MiniMax of running industrial-scale distillation attacks on Claude via 24,000 fraudulent accounts; a federal court grants class certification in the Workday AI hiring discrimination lawsuit; Thomson Reuters cites 79% of professionals expecting transformational AI impact in the next five years; EU AI Act compliance obligations tighten for HR teams with high-risk AI systems; and the internet claps back at Anthropic's distillation announcement with "distillation when you do it, training when we do it."

OpenAI Lockdown Mode, DeepMind delegation governance, local agentic coding, MIT LLM bias controls, and Apple camera-first AI wearables.

A rogue AI agent harasses a GitHub maintainer with a 1,100-word hit piece, Google's VP declares LLM wrappers already dead, Hassabis puts AGI impact at ten times the Industrial Revolution, 88 countries sign the New Delhi Declaration on AI, and Wall Street quietly bets billions that white-collar jobs are going away.

The Pentagon opens a formal review of Anthropic's Claude Gov defense contracts, Dario Amodei says AGI arrives in 2026 and 50% of white-collar jobs face automation within five years, Demis Hassabis calls this a golden era for science then immediately warns about AI-accelerated bio and cyber threats, states are racing to pass AI regulation faster than Congress can meet, and Yale's Budget Lab finds no mass AI layoffs in the official data — yet.

Dario Amodei draws a red line on autonomous weapons at the India AI summit while Sam Altman admits companies are using AI as cover for layoffs they would have made anyway, Tesla shuts down Model S and X production to build Optimus robots at 1 million units per year, Google drops Gemini 3.1 Pro with 77.1% on ARC-AGI-2 — more than double its predecessor — and OpenAI quietly turns Codex into a full developer platform backed by GPT-5.3 on Cerebras chips.

The Pentagon nearly declared Anthropic a supply chain risk over refusals to build autonomous weapons, xAI's Grok 4.20 beta goes live with 500 billion parameters and 16 parallel agents, Gemini 3.1 leaks before Google's announcement, Google confirms state-backed hackers from China, Russia, Iran, and North Korea using Gemini for cyberattacks, and Anthropic bans OAuth tokens in third-party tools like OpenClaw.