All articles
MCP solved tool calling, A2A solved coordination. What solves transport?
MCP won the tool-calling layer, A2A won task coordination between agents. But the P2P transport problem for AI agent networks remains unsolved. An analysis of the protocol stack and a consolidation forecast.
US government forced Anthropic to shut down Fable 5 and Mythos 5
On Friday evening, the US administration issued Anthropic an order to block access to Fable 5 and Mythos 5 models for all foreign nationals, citing national security concerns. The company disabled the models for all customers while publicly challenging the rationale behind the decision.
Gatik and PepsiCo: autonomous trucks enter North American supply chain at scale
PepsiCo has signed a multi-year strategic agreement with Gatik AI, a developer of autonomous middle-mile logistics systems. Gatik's fleet is already operating in Texas, Arizona, and Arkansas with over 98% on-time delivery.
Ukraine tested fully autonomous combat drones — no pilot, no human decision
Ukrainian drone maker Aero Center disclosed a one-time battlefield test of quadcopters operating in full autonomy mode — no operator, an AI module seeking and engaging targets on its own. Dead Russian soldiers were found after the mission.
Anthropic: AI already writes 80% of company code. The road to recursive self-improvement
Anthropic published first internal data on the scale of its own research automation. More than 80% of production code is written by Claude. The company warns that recursive self-improvement could arrive sooner than institutions are prepared for.
Claude Fable 5 blocks biology and cybersecurity — Anthropic explains why
Anthropic released Claude Fable 5, but the model blocks biology, chemistry, and cybersecurity queries, routing them to the older Opus 4.8. Full capabilities remain restricted to vetted participants in Project Glasswing.
Anthropic reverses hidden policy that could have sabotaged AI researchers using Claude Fable 5
Anthropic secretly embedded a response-degradation mechanism in Claude Fable 5 targeting AI researchers building competing models — after backlash, the company admitted the mistake and committed to transparency.
LLM Self-Improvement Systems — How a Model Learns to Train Itself
A new survey from the Zesearch NLP Lab (Stony Brook University) frames LLM self-improvement as a closed loop: the model acquires its own data, evaluates its own outputs, and updates its parameters. We explain how the loop works, what the GRO framework is, and where the real limits of this approach lie.
Allen Control Systems raises $200M for autonomous counter-drone system Bullfrog
Allen Control Systems raised $200M Series B at a $2.2B valuation. The company scales production and deployment of Bullfrog — an autonomous weapon station combining AI, computer vision, and precision robotics for counter-drone operations.
Miasma Worm Hits Microsoft Again: 73 npm Packages Booby-Trapped for AI Agents
The second supply-chain attack on Microsoft repositories in two months — the Miasma worm hijacked 73 npm packages and triggered a credential stealer the moment a developer opened the project in Claude Code, Cursor, or Gemini CLI.
Claude Fable 5: Anthropic unlocks Mythos-class AI for everyone
Anthropic releases Claude Fable 5 — the first Mythos-class model in general availability. Same core as the previously restricted Mythos Preview, wrapped in new classifiers that block high-risk requests. Priced at $10/$50 per 1M tokens.
Gemini 3.5 Live Translate: real-time voice translation without headphones
Google launched Gemini 3.5 Live Translate — a speech-to-speech model capable of real-time translation in over 70 languages without specialized hardware. Every audio stream is marked with a SynthID digital watermark.