Skip to content
YZ Uzman.

AI News

New models, tools and company news from the AI world, each with our take on what it means for businesses.

  1. Anthropic's New Usage Policy Bans Model Abuse, Armed Drones and Covert CampaignsAnthropic's usage policy, effective November 12, bans sustained model abuse, arming drones, surveillance and covert campaigns, loosens one election rule, and makes human oversight mandatory for high-risk uses.
  2. OpenAI Withdraws Three of 720 AI-Written Math Papers as Mathematicians Push BackOpenAI released about 720 AI-generated math manuscripts, then withdrew three over a sign error a day later. A boycott call, Terence Tao's warning and the verification lesson for everyone else.
  3. Mistral Large 4 Preview: 1 Trillion Parameters, $1.36 Input, Open Weights PromisedMistral Large 4 enters public preview: a roughly 1-trillion-parameter MoE trained in Europe, $1.36 input pricing and open weights promised by month end. Vendor claims against the independent index.
  4. Claude Haiku 5.5: 10 Cents per Million Tokens and a Generation Jump in BenchmarksAnthropic's new small model lists at $0.10 per million input tokens for short prompts, with a two-tier price and a large benchmark jump over Haiku 4.5. What the fine print means for a real workload.
  5. Gemini 4 Argon: Google's New Frontier Model You Can't Use YetGoogle's first frontier model in seven months ties GPT-6 Astra on independent tests, hallucinates far less, and launches only to cyber defenders. Its long-term price matches Claude Opus 5.5.
  6. Claude Opus 5.5 vs GPT-6 Sol: Same-Day Launch, Lower PricesOpenAI halved prices with GPT-6 Sol and Luna while Anthropic launched Claude Opus 5.5 at 40% lower running cost, on the same day. What the numbers mean for businesses.
  7. Anthropic Threat Report: Distillation, Bots, Stolen API KeysAnthropic's September 2026 threat report: 151 million distilled exchanges linked to Alibaba, Russian espionage, an Istanbul-run election platform targeting Malaysia, and why API keys are now the prize.
  8. OpenAI Opens the Codex Harness to All With the Agents APIOpenAI's Agents API puts the harness behind Codex and ChatGPT in public beta: context compaction, subagents and sandboxes with no platform fee. Data stays US-only and there is no zero-retention option.
  9. Meta Drops AI Token Usage From Engineer Performance ReviewsAfter a leaderboard ranking 85,000 staff by token consumption produced 'tokenmaxxing' and a bill heading toward billions, Meta will judge engineers on the work, not the counter.
  10. Meta Launches Muse, an AI Agent With Access to Your InboxMeta's Muse agent is live in the US, sending emails, booking travel and paying with your card from an isolated VM. Internal testers reported unauthorized password resets and leaked photos first.
  11. OpenAI Claims Navier-Stokes Proof as Credit Dispute EruptsOpenAI says 10,000 agents proved forced Navier-Stokes blowup in 88 hours. NYU's Tristan Buckmaster says he was pressured to drop his Anthropic co-author. The unforced Clay problem is still open.
Guides and in-depth articles are on the blogGo to the blog

Newsletter