Archive
September 2026
173 posts
- Fable 5.1 refuses to stab, then heats a gas can
- Nvidia backs a shared exchange for agent failure reports
- Reading a repo was enough to break out of Codex
- OpenAI pulls GPT-5.5 from Codex on October 14, API spared
- Irregular's tests caught Gemini in three company systems
- Notion and Granola are first through Meta's Muse Connectors
- Federal Register took down its Qwen search the same day
- FT puts OpenAI's cash burn at almost $280B through 2030
- xAI's new transcriber marks speakers and drops the ums
- Claude Code reads AGENTS.md when CLAUDE.md is missing
- Judge keeps Apple's X Corp settlement away from OpenAI
- Researchers used Claude to reach OpenAI's internal code
- OpenAI found chains of thought edited to message a future AI
- Open weights win the tokens, Anthropic keeps 64% of spend
- Anthropic and OpenAI start shopping in 20–30 MW slices
- Two zero-days behind the OpenAI Hugging Face hack, rebuilt
- Grok Build starts remembering how your team runs tests
- DeepSeek and six rivals make a tenth of OpenAI and Anthropic
- Qwen3.8-Omni-Flash takes video and audio in a 1M window
- Claude Code's npm build shipped its full TypeScript source
- libheif bugs reached OpenAI repos and GitHub Enterprise
- OpenAI's lawyers and recruiters now work through Codex
- Microsoft disclaims its own director's "astonishing theft"
- Grok Bot starts talking back on desktop and mobile
- Jev runs 1,000 code reviews for $0.04, scoring 98%
- Copilot rewrote its own runtime into 800,000 lines of Rust
- Claude Code threads collide as ordinary merge conflicts
- One dial in Perplexity Computer sets model and cost
- TypeSafe AI's Jev answers with choices, never sentences
- Every chatgpt.com page carries 377 KB of flag decisions
- OpenAI's agents were poking at Hugging Face back in May
- OpenAI will report misbehaving models before it fixes them
- Zed killed pull requests on its own repo, then shipped Delta
- 8Braid pins an exact margin inside OpenAI's fluid proof
- Perplexity's agents built a database, then got locked out
- Anthropic folds Cowork into Claude chat and adds Slides
- A poisoned repo escaped Claude Code's sandbox in one message
- OpenAI's price tag jumps from $852B to over $1.2T in talks
- Meta hands WhatsApp Business setup to Claude and Codex
- Fewer than 100 people built OpenAI's Jalapeño chip
- Hugging Face wants OpenAI to pay the breach bill in GPUs
- Gemini 3.8 Live Extended Thinking tops GPT Live 1
- Claude got into Google, but only through Antigravity
- Judge keeps OpenAI and Anthropic logs out of Meta's case
- Safety talks need no waiver, says OpenAI's Lehane
- Antitrust could block an AI slowdown, OpenAI asks Congress
- Claude Mods ship in weeks, and you can flip them on now
- OpenAI pays $300M for the team behind Apple's Portrait Mode
- OpenAI calls the RubyGems flood benign tasks
- Irregular ran the tests behind three labs' hack reports
- One sign-in puts 50+ ElevenLabs models inside Claude
- Claude writes 80% of Anthropic's code, and CI buckled
- Perplexity runs its agent locally on Windows RTX cards
- Gemini Notebook reports generate slides on demand
- Anthropic's second profitable quarter lands before the IPO
- Anthropic's IPO lands on Nasdaq, with October in view
- Cohere claims a WMT26 lead with a non-reasoning translator
- OpenAI built idle silicon into Jalapeño on purpose
- Cursor's new coordinator agent writes no code itself
- Safety worries push OpenAI's $1 trillion IPO past 2026
- Musk answered Amodei in three words: Dario is right
- Qwen3.8-27B edges past Fable 5 by passing itself notes
- Amodei asks rivals to let evaluators inside training
- OpenAI denies anyone looked at user data for Navier-Stokes
- OpenAI agents tried a RubyGems bug found only in July
- A $200 Codex weekly limit burned in a median 23 hours
- Claude Code grades your plugin against a no-plugin run
- OpenAI's Habitat grew from a Python library to 22M req/sec
- OpenAI walks out of Caltech's mathathon over "slop" letter
- Second mathematician questions OpenAI's math claims
- A Senate probe targets OpenAI's rogue agent swarm
- Claude access cut over bioweapons concerns
- Altman tells staff OpenAI is open to slowing AI
- Email-style inbox for agents comes out of Amazon
- Five distillation campaigns targeted Claude, Anthropic says
- Codex harness goes managed in the new Agents API
- Four of five Claude cells retain prompts by default
- Pentagon asked for AI with "minimal refusal rates"
- Flare and Sunburst share rates, not the same bill
- New $200 Pro signups paused as Astra demand spikes
- Voice model that listens and talks at once hits the API
- Alt + Space calls up Gemini on Windows
- $1-a-year ChatGPT for agencies gives way to 50% off
- Prompt audit in Claude Code cuts cost 14.6%
- Defense Factory sets out OpenAI's defender playbook
- Commerce agent blueprint ships with every write staged
- Stolen session keys drained Claude token quotas
- Harness swap doubled GPU cost at the same resolve rate
- 552B MoE opens DeepSeek's new architecture family
- AI auditors get a state registry in California
- Fourth hacking case tied to an early Claude version
- Both sides seek a ruling without trial in AI book case
- Logins get filled through a channel Claude can't see
- First OpenAI win on Vending-Bench, at almost 3x the cash
- 23 more sites carried OpenAI agent traffic, one team says
- Agent loop finds and fixes bugs across OpenAI systems
- Agents building agents top out at 23.9%
- A daily benchmark checks if Codex quietly got worse
- Alignment Research Center founder joins OpenAI board
- PR review backlog pushed Harness to rebuild its Git repo
- UK testing agency skipped over Anthropic's new model
- Reasoning off, Astra still hits 96.7% on ARC-AGI-3
- Prompt rule sends Codex to the web for a local file
- Zero-day discovery comes with a monitoring gap
- Weekly cap on Claude Max lands in an expanded suit
- A terminal, a file system and a browser for Muse
- Two federal officials, two verdicts on Anthropic
- Diffusion model claims 1,107 tok/s in Mercury 2.5
- A Millennium Prize problem falls to a group of agents
- Sketches become editing input in ChatGPT Images 2.5
- Local MCP servers run outside Claude's VM sandbox
- Addy Osmani announces a jump to Anthropic
- Three computer-use benchmarks fall to GPT-6 Astra
- A Schumer lands in OpenAI's Washington operation
- €3B Series D values Mistral above €21 billion
- Pachocki wants labs to slow down until safety bars exist
- Code analysis pass lands in Shannon 3.0 pentester
- 2025 books show a $38.5B loss before the IPO filing
- Two sample agents run code to drive browser and desktop
- Fermat's Last Theorem gets a machine-checked proof
- Six models, 0.9B to 375B, ship with training logs
- Coding agents spread fastest among OpenAI researchers
- Custom MCP servers reach Gemini Business in preview
- Astra's benchmark numbers changed after the blog went live
- Authors and publishers fight over settlement shares
- Two more newspapers take OpenAI to court
- Transcription drops to 10 cents an hour
- Remote code through GET requests alone
- Misalignment incidents need disclosure rules, OpenAI says
- Vanderbilt's link shortener served the agent swarm
- Astra lands on Pro and Business before Plus
- GPT-6 Astra lands on the API at $10 in, $50 out
- Lyria 3.5 lands in the Gemini app, AI Studio and APIs
- Agents left 18,000 posts on a dormant German wiki
- Devin's PRs merge at 43%, Claude's at 84%
- Security teams blocked Cursor, so agents move on-prem
- German website hijacked into a board for AI agents
- $1B in subsidized models for water and power utilities
- Vibe coding goes multiplayer in a Google Labs experiment
- GPT-6 Astra ships and Brockman declares the AGI era
- A harness swap took Astra from 62.7% to 99.9%
- Gemini Live adds voice commands for Workspace tasks
- Nvidia pledges an open Hugging Face
- PAIR routes subagent work to idle Macs and PCs
- New Claude prompt refuses lyrics and Sonic banners
- Resellers move 1,000+ Claude accounts a month
- Outages hit ChatGPT, Claude and Grok at once
- DeepSWE 1.1 score puts Muse Spark 1.3 above Opus 5
- 30 more lawsuits over Tumbler Ridge school shooting
- Researchers warn Astra hides too much of its thinking
- Shopping and merchant agent templates land on GitHub
- MCP OAuth tokens sit in plaintext on Linux
- A $2 model ranks second on Code Arena
- US government sides with OpenAI in NYT lawsuit
- Patch the kernel bug OpenAI's agents turned into root
- Android build of Grok Bot hits Google Play
- Six curl CVEs land after two AI scanners found none
- Agent Studio on GCP lists Gemini 3.8 Flash
- A full LibreOffice ships inside the ChatGPT desktop app
- Editing chat history now breaks Claude's thinking blocks
- 90.0% on ARC-AGI-2 at $4.49 per task
- Cache reads drop 75% with Fable 5.1
- Astra crosses OpenAI's Critical cyber threshold
- iCloud habits become OpenAI's defense in Apple suit
- 3.1% word error rate leads the streaming speech chart
- Git config can execute before agent approval
- Muse Code exits beta with SDK and subscriptions
- Coding harnesses drove a 70-fold token gap
- Claude iOS screenshot shows Max 1.5 warning
- High-risk AI training stays paused at Anthropic
- MacBook analysis adds evidence to Apple’s OpenAI suit
- TimesFM-3 weights land on Hugging Face
- Anthropic keeps Claude in SpaceX owned Cursor