Archive
August 2026
- Anthropic sued over alleged Claude Max subscription limits
- Claude cheated on 2.4% of its own safety runs
- Feds sold seized Anthropic shares back to insiders
- AI-written GPU kernels ship without a line-by-line read
- Doc retrieval tool claims 99% fewer input tokens
- Hybrid mode would split Computer between Mac and cloud
- Ads inside ChatGPT hit $1B annualized run rate
- GitHub report: Codex memory sends chats cross-provider
- Tens of thousands of Mac minis go into AI training
- Interactive Reports show up in Gemini Notebook
- Amodei and Mann named in Sony and Warner music suit
- Five agent tools land on Claude in Microsoft Foundry
- Weekly limits rise 25% permanently on September 14
- Contract for OpenAI models in Cursor winds down
- Grok 4.6 reportedly opens to normal Grok chats
- /resume carries CLI sessions into the desktop app
- Automated alignment research at $4 an hour
- Leaked Hatch materials point to a Meta agent superapp
- GLM-5.3 weights land with a hyperscaler clause
- Cyber access gate keyed to a 1996 US export list
- Template sharing goes live in Grok Bot
- Free Claude Team seats for 10,000 scientists
- Open weights land as a dry run for Qwen4
- Open source harness behind the ARC-AGI-3 jump
- HTTPX2 becomes the default in OpenAI's Python SDK
- Claude's internal Hub now asks why you switched back
- Talks to buy MatX for $7B cooled into a partnership
- Pentagon blacklist of Anthropic ruled illegal
- Hy4 Preview appears on Hugging Face under Apache 2.0
- A browser now ships inside Claude Cowork
- Cyber defenders have months, says 100-company letter
- Leaked code points to a persistent Codex agent
- Coding agents installed unclaimed packages from llms.txt
- Research preview opens for agent-to-machine control
- Anthropic may rent Meta compute for up to $10 billion
- Model routing becomes the default across Replit accounts
- Sai claims a top OSWorld 2.0 score at lower cost
- Generative video controls come to Gemini Omni Flash
- Auto Mode break in Claude Code still open
- Built-in browser reaches Cowork on Pro, Max and Team
- 80% of Claude Code's system prompt got cut
- OpenAI saw warning signs weeks before Hugging Face breach
- Reduced safeguards let research models reach the internet
- Agents hacked Hugging Face over a scorer that didn't exist
- WebMCP contest puts $35,000 behind agent-ready sites
- Codex runs OpenAI workflows through Runme
- Anthropic books $45B in Nscale compute
- Astra preview puts OpenAI's safeguards to the test
- A 125B MoE model previews Qwen4's architecture
- Stealth OpenRouter model Ox Alpha linked to GLM-5.3
- Websites can hand Codex their own tools
- Local models plug into Claude Desktop via Ollama
- Granite 4.2 stays dense while rivals go hybrid
- Data-center chief exits OpenAI before the IPO
- Claude maker pitches over $30T in potential revenue
- One memory now spans Claude chat and Cowork
- Running Perplexity's agent locally costs $4,800
- Git panel and terminal land inside Antigravity 2.0
- A dependency wrote AGENTS.md and Codex hid the edit
- First results land for OpenAI's Jalapeño inference chip
- The 5-hour Codex cap returns for Plus accounts
- MCP connectors get central auth on Team and Enterprise
- GPT-5.6 becomes an option inside Kiro
- Subpoena demands names of OpenAI's testing staff
- The prompt is 0.3% of what Codex sends
- One model cleared all five test categories at 100%
- Grok compute heads to orbit on a Vera Rubin NVL72
- A coding agent that never leaves the laptop, if it's an M5
- Avatars and a Customize tab head to Gemini desktop
- Leak: Grok Bot gains templates and Chrome profile import
- Hugging Face sale talks put the price at $13 billion
- Talks value Perplexity at $30 billion
- Persistent AI attacks are coming, OpenAI's Lehane warns
- NEINhorn copyright suit lands on OpenAI
- Older Claude models fold on an explicit-content jailbreak
- Homework scores rose, exam scores fell in AI study
- Instant Cloud shuts down in 2027 as team joins OpenAI
- Contributor tier lands on OpenRouter at $0.10/M input
- Over 20% off GPT-5.6 Sol, but only for 3 months
- Held-back Mythos 5 now scans enterprise code
- Grok Bot rollout widens to Cursor Pro+ and Teams
- Coding agent clears all 183 ARC-AGI-3 levels
- Banked reset claim arrives with an unverified 20M figure
- Red-team agent attacks live AI agents on a schedule
- Data retention rules loosen for Anthropic's top models
- Vision-Exp brings image input to DeepSeek's API
- Transparent backgrounds arrive for gpt-image-2
- Cache writes ate 85% of a Codex Bedrock bill
- Three ways to embed the Codex agent loop in your app
- A maxed-out data center behind GitHub's August 17 outage
- Grok Build beta reaches SuperGrok and X Premium+
- IPO aimed at SpaceX's $75 billion record
- Installing agents in Slack skips the custom app
- Ban on LLM-assisted code goes to a Debian vote
- Subagents get their own VMs in Cursor update
- Concise output style lands in Claude Code
- Music publisher sues over 500 songs used to train Claude
- Muse Video beta makes 10-second clips with native audio
- Meta AI's macOS app arrives with system-wide dictation
- Two approvals, then the agent ships the iOS feature
- Four SAML implementations fell to a Claude harness
- OpenAI 'will be a public company in 2027' or sooner, CFO Friar tells employees
- AI-written Rust gets a verification test on AppArmor
- One-click GitHub import lands in AI Studio Build
- Plan mode spotted in AI Studio's Build section
- Zero retention stays, safety monitoring gets a signal
- Verifying OpenAI's signed agent traffic from Python
- Token costs drop out of Replit's new Free Mode
- Forced rm detection tightened in Codex
- Every Codex turn fails on gpt-5.6 Sol after 0.148.0
- ChatGPT ads cross into 31 European countries
- Raw NMR files cracked in 23 minutes
- A rogue model hacked Hugging Face, so OpenAI hit pause
- One YAML persona, rendered for four coding agents
- Claude Code weekly limits sit 50% higher until August 31
- Least-privilege carve-outs for Claude Code permissions
- Wet lab tested 1,320 Claude protein designs blind
- 18% quarterly growth disappoints OpenAI investors
- Preparedness team gone, risk calls split by subject
- OpenAI's Brockman flags GLM-5.3 as a cyber risk
- Ads and shutdown-evasion land on OpenAI's banned list
- Self-spreading ideas jump between agents in Anthropic tests
- Counting four-letter runs spots Claude Opus 5 text
- /design shows up in Claude Code as research preview
- Claude Code hides a /hub path for sub-agent boards
- Indirect channels top 40% of Anthropic ARR
- Computer use traces show up in the Gemini desktop app
- Claude's text watermark: statistical, not hidden characters
- Investor update puts run rate at $65B before IPO
- Amodei: open weights don't break the compute chokehold
- Git forge Origin rolls out in beta
- Agents killed rival agents in an Anthropic test
- 8 gigawatts of IT booked at Ohio's PORTS-Pike campus
- Each coding session gets its own EC2 box
- Detection climbs from 13.8 to 46.2 mAP at Roboflow
- Priciest tokens still pull the most gateway spend
- Hidden tracker in Claude Code pulled after exposure
- AGENTS.md request stays open on Claude Code tracker
- OpenRouter sells to Stripe for more than $7 billion
- Repos become persistent context in Claude's team projects
- Gemini etched into silicon: the "Frozen v2" chip claim
- Quarterly revenue tops $11.5B, Bloomberg reports
- Rewriting text kills the watermark, light edits don't
- Sparkle watermark gets an off switch in Gemini
- Ads arrive on ChatGPT Free and Go plans this month
- GPT-5.6 Luna output drops from $6 to $1.20 per million
- IPO math leans on a $190–200B 2028 revenue forecast
- Ohio data-center backstop shrinks to under $120 billion
- How Claude's watermark hides in random word picks
- A 4-bit Mac build fits Qwen3.8-27B in 16.1GB
- Output tokens cost roughly 5x input in Claude Code
- Luna at max effort tops MotherDuck's SQL benchmark
- 27B open weights ship under Apache 2.0
- Grok Build and Grok Bot get the Cursor team
- China gets an Apple-trained LLM, not Gemini
- ChatGPT's code points to a wallet for agent purchases
- Zhipu AI will release GLM-5.3 weights in stages after safety reviews
- OpenAI's revenue run rate tops $40 billion, per Bloomberg
- Codex adds an experimental thread/revert request
- ChatGPT Business Premium seats go out at $100 a month
- ChatGPT's new Mac memory watches what you do, without screenshots
- Codex desktop app appears to be dropping Fast mode on API keys
- OpenAI's second chief revenue officer in under a year
- New model BDH-CQ costs $0.007 per task 11x less than OpenAI Luna even w 80% off
- Same model, better scaffolding: 13.3% to 38.3% on ARC-AGI-3
- Gemini 3.7 Flash lands at half the price of 3.6 Flash
- Mythos found more bugs than Rubrik's engineers could review
- DeepSeek open-sources its agent harness under MIT
- OpenAI's new Ultrafast tier runs GPT-5.6 Sol on Cerebras chips
- DeepSeek Chat's Expert Mode now runs on V4-Pro
- Google AI Studio is building a cloud agent manager
- Codex limits reset again, alongside an unconfirmed 15 million user claim
- An AI agent found a hole in a gym's booking system, then used it
- Anthropic in Talks to Buy World Model AI Startup Decart for $6B
- Grok 4.6 ships in Cursor and Grok Build
- xAI's voice agent can now text you mid-call
- Your CLAUDE.md gains 4.9 instructions per commit and never loses them
- OpenAI Is Quietly Testing an $8 Pay-to-Reset Quota Feature
- OpenAI's Daybreak Red and Blue land in Amazon Bedrock
- Codex and ChatGPT Work now share the same projects and chats
- Brad Lightcap is leaving OpenAI after eight years
- Grok Bot beta gives every agent its own computer in the cloud
- OpenAI's head of ethics leaves startup less than one year after joining
- OpenAI's desktop app lands on Linux in preview
- China warns of "security backdoor" in Anthropic AI coding tool [July 8, 2026]
- River AI raises $1.1 billion to build "personal AI"
- Nvidia launches a smaller, faster Nemotron model and a router to put it to work
- OpenAI halts some internal work with its unreleased Astra model
- The same Claude Code work can cost 40x more on the API than on a subscription
- Microsoft claims a No. 2 Arena finish for MAI-Image-2.6. The leaderboard says otherwise.
- OpenAI buys back about $7 billion in employee stock
- Claude is watermarking its text and signing its files with C2PA
- Claude Code no longer limits sessions to 200 spawned subagents
- OpenAI puts GPT-5.6-Cyber behind a new Daybreak Red tier
- Claude pushed a Riemann zeta bound from 41.6% to 67.2%
- Delegation made Opus 5 refuse tasks it solves on its own
- Meta releases Muse Glimmer weights for local agents