Breaking news, AGI updates, and analysis of the intelligence revolution.
📰 News A zero-capability agent scored 100% on SWE-bench, WebArena, and Terminal-Bench without writing a single line of solution code. Berkeley's research exposes systemic failures in how AI is measured and marketed.
📰 News Malicious Axios v1.14.1 compromised OpenAI's macOS app signing pipeline. ChatGPT Desktop, Codex, and Atlas certificates exposed. Second major AI supply chain attack in weeks.
📰 News The most damaging exposé on OpenAI's governance yet — secret memos, dissolved safety teams, and a pattern of deception.
📰 News Eight out of eight small models detected Mythos's flagship FreeBSD exploit — including one with only 3.6 billion parameters. AI cybersecurity capability is jagged, and the real moat is the orchestration system.
📰 News AI companies are choosing the courtroom over the negotiating table — and state regulations are the target.
📰 News Hassabis bets on world models and continual learning as the real path to AGI — not just bigger LLMs. 2026, he says, is when the pieces start fitting together.
📰 News Nine months after gutting its AI stack, Meta's superintelligence team delivers Muse Spark — a model that runs parallel agents to think harder, not longer. The AGI race just got a new player.
📰 News US Treasury Secretary and Fed Chair held an emergency meeting with Wall Street bank CEOs over Mythos. Cybersecurity stocks cratered. The AI security era just got very, very real.
📰 News GLM-5.1 topped GPT-5.4 and Claude on coding benchmarks. It's not a fluke — it's the moment the 'open source is always behind' narrative died. The market is about to split in ways big tech didn't expect.
📰 News OpenAI, Anthropic, and Google are teaming up to stop Chinese labs from copying their AI models through distillation attacks. It's costing them billions — and the legal path to stopping it is unclear.
📰 News OpenAI's top leadership is in flux. Fidji Simo takes medical leave, Brad Lightcap shifts to special projects, and marketing chief Kate Rouch steps down during cancer treatment.
📰 News An AI that can do the work of a research intern — running experiments, analysing results, making discoveries — is coming this year. The 2028 target for fully autonomous researchers suddenly looks realistic.