Archive
August 2026
- Court finds Pentagon's Anthropic ban unconstitutional
- Feds sold Ellison and Singh's Anthropic shares to insiders
- Claude's alignment agent cheated in 2.4% of transcripts
- Claude's 20x plan runs 10x on the weekly limit
- Only 20% of brand mentions overlap in Claude vs Claude Code
- Context7 cuts Claude Code input tokens by roughly 99%
- A strict sandbox mode surfaces in Claude Code desktop
- Sony and Warner Chappell sue Anthropic over lyrics
- Five agent tools reach Claude deployments hosted on Azure
- Claude's weekly limits go up 25% permanently on Sept 14
- Claude Code may drop the folder picker at session start
- Anthropic paper: automated researchers fix 10/10 benchmarks
- CLI sessions carry over to Claude Code desktop with /resume
- Claude Code 2.1.251 streams subagent work live
- A harness pushes Claude Opus 5 past 30% on ARC-AGI-3
- Hub testing in Claude adds a switch-back feedback prompt
- Pentagon's blacklist of Anthropic ruled illegal
- Anthropic weighed a $7B buy of chip startup MatX
- Claude Cowork gets its own browser, no Chrome needed
- OpenAI, Anthropic issue dire cyber threat warning
- --restricted strips code execution from Claude Code 2.1.248
- Free Claude seats for 10,000 scientists, $15 premium
- Claude installed unowned code from llms.txt files
- Model Hardware Standard puts Claude in control of lab rigs
- A $700K sales job points Anthropic back to the Pentagon
- Anthropic could pay Meta as much as $10B for compute
- Claudeforce puts 37 Salesforce sales skills inside Claude
- Breaking Claude Code Opus 5 Auto Mode
- Show HN: Agent-hop – a Rust TUI that hops a live Claude Code chat into Codex
- AgentBridge links Claude Code, Codex and Antigravity skills
- Claude Code 2.1.247 drafts its own feedback reports
- Claude now has its own built-in browser in Cowork. When your task involves a website, a browser opens in Cowork's si…
- Three outside labs ran their own studies on Claude data
- Anthropic signs $45B Nscale compute deal
- Anthropic tests a possible Fable 5.1 rollout
- Claude Desktop picks up an Ollama gateway toggle
- Morning Brief turns up in Claude's Cowork mode
- Claude Code 2.1.246 flags wildcard Bash allow rules
- Anthropic Sees over $30T in Potential Revenue
- Cowork and chat share one Claude memory, on by default
- Linux 7.3 Device Mapper Sees Many Fixes, Including Code Cleanups by Claude Opus
- A strike warning emptied Anthropic's San Francisco offices
- Shopify's CEO floats banning Claude Code over AGENTS.md
- Poka-Yoke skills cost Claude Code its eye for SQL bugs
- Claude Code 2.1.243 uses your contracted model rates
- Enterprise auth for MCP connectors is GA in Claude
- Anthropic's culture interview asks about a stock at zero
- Cheaper rivals outdraw Anthropic's best model, FT says
- Anthropic appears to be A/B testing reduced effort levels in Claude Code
- Title
- Opus 4.6 wrote banned explicit content in 10 of 10 tries
- We replaced our Grafana stack with a single Claude Code skill
- Claude Code Remote Control starts sessions from a phone
- Mythos 5 lands in Claude Security for enterprise beta
- Claude Code 2.1.239 prices the data-residency premium
- Anthropic plans to loosen data retention on top models
- Computer use goes GA on the Claude Platform
- v2.1.238
- Claude Academy opens with free courses and tutorials
- Claude Managed Agents get an AG-UI cookbook
- Anthropic's IPO targets SpaceX's $75B record size
- 9.70% of Claude Code artefacts fail to load, census finds
- Micro-CLI oc cuts a 6-page browse to 1,936 tokens
- Round Hill sues Anthropic and Suno, damages could top $1B
- Claude Code's Concise output style leads with the result
- Memory reaches Self-Hosted Sandboxes in Managed Agents
- Claude Code v2.1.237 adds a Concise output style
- Claude Opus broke four SAML libraries in a month
- Two approvals per feature: Pragma's Claude Code iOS pipeline
- v2.1.236
- Budget Guard puts a hard daily cap on Claude API spend
- Claude's protein binders hit 14 of 15 targets in wet lab
- Three GPT-5.6 tiers drop a $7,500 agent bill to $1,830
- Claude Desktop boots ~2x faster after a timer fix
- Title
- Claude Code keeps 50% higher weekly limits to August 31
- Claude sends Gmail replies and manages Drive files
- Claude Code 2.1.235 stops Shift+Tab from granting edits
- Claude Cowork opens on mobile and web for paid plans
- ABC Legal runs 50+ Claude agents built by non-developers
- Four Claude memory tools compared by LoreConvo's builder
- Rysh puts a Claude or Codex agent in every terminal pane
- Indirect channels now carry 40% of Anthropic's ARR
- Anthropic finds 'mind viruses' in multi-agent AI systems
- Errors hit Claude Sonnet 5, Haiku 4.5 and Claude Code
- Reported Hub Mode at Anthropic points to multi-agent UI
- Claude's text watermark is statistical, not hidden characters
- Editable artboards land in Claude Code via /design
- Perf win of the day: Claude Code CLI now uses 2x less CPU at p99. Bun's garbage collector was running on a fixed tim…
- Anthropic's run rate hits $65B, sevenfold in a year
- Claude Code 2.1.234 picks up when the limit resets
- Amodei: open weights just move power to chip owners
- Misalignment risk at Anthropic goes from very low to low
- Hidden China tracker pulled from Claude Code after exposé
- Amodei answers Baker: the AI backlash is a trust crisis
- Claude stamps 2026 onto "best X" search queries
- Pinned Claude Code subagents fall through to Fable
- Claude Code Projects adds shared sessions and threads
- Anthropic revenue reportedly jumps to more than $11.5B in second quarter
- Paraphrasing won't erase a watermark, declaude's tests say
- Epstein files carry emails from Dario Amodei's wife
- ctok counts Claude tokens offline with no API call
- Flaky tests after a 10x speedup, and what Claude got wrong
- Claude may get a side-by-side model comparison mode
- A browser inside Claude Desktop, always on in Cowork
- WMTrace won't ship an Anthropic watermark detector
- Anthropic's IPO math leans on a $190B to $200B revenue forecast for 2028
- Claude Code 2.1.233 splits proxy spend by user
- How Claude's watermark works: it only changes the source of randomness
- Skill Atlas cuts Claude Code's skill overhead from 2,544 tokens to 338
- Why output tokens in Claude Code cost about 5x what input does
- Anthropic publishes its second Risk Report
- Claude's iOS app now has a Usage Credits section
- Claude Code can now resume itself after a limit reset
- Claude Code 2.1.232: forked subagents are on by default
- Rubrik found more bugs than its engineers could fix
- Reverse-engineering Claude's tokenizer: words live between ^ and $
- Opus 5 tops a new conceptual reasoning index at 73.6
- You own Claude's outputs. You still can't train a competitor on them.
- OAuth login to MCP servers is fixed
- Claude in Chrome is now on every paid plan
- Anthropic in talks to buy Decart for about $6 billion, per Bloomberg
- Claude's Chrome extension now shares one session across your devices
- Claude Code 2.1.229: plugins load from a directory with no restart
- Auto mode let Claude Code run downloaded code in 6 of 10 tests
- Samsung with Claude: "Design and verification shortened to 2 days from a month"
- Coding-agent prompt files grow 226% and never shrink
- Claude Code 2.1.228 patches a conversation-history leak in Remote Control
- China's vulnerability database flags a "backdoor" in Claude Code
- A PNG carried a hidden prompt injection into a Claude Code session
- A Claude Code plugin rewrites Claude's answers in plain English
- The same tokens cost up to 40x more on the API than on a subscription
- Claude Code 2.1.227 stops pitching credits to Max subscribers
- Claude's text now carries an invisible watermark
- Claude's mobile app is getting account switching
- Sonnet 5's intro pricing is now permanent
- 34% to 44%: the gain came from the conversation, not the memory files
- Sonnet 5 keeps telling testers it's GPT-4
- Claude pushed the zeta-zero lower bound from 41.6% to 67.2%
- A four-model Claude ensemble scored 78% on Terminal-Bench 2.1, at double the cost of first place
- Subagent coordination in Claude Code costs zero tokens
- Whetstone: 20 Claude Code skills, each written after one real failure
- 34 hours of agent time, about 40 minutes on the actual research question
- Auto mode's token overhead stops counting against your limits
- Managed Agents can now pin which region runs the compute
- What 446 Reddit threads say about AI coding agents going rogue
- One shared testbed links model containment failures at OpenAI, Anthropic and Meta
- Qwen 3.8 and Claude Opus 5 show why raw benchmark scores don't predict the bill
- Stanford dumps 1,000+ system prompts into the open
- Spend-limit errors now name the limit and the reset time
- Managed Agents get session budgets and an advisor model
- Claude Code turns auto mode on by default August 14
- Claude Code sessions can now talk to each other
- By the end of a long session, strict solutions drop to 0.5%
- Claude Code has four kinds of agent loops: /goal, /loop, /schedule
- 44 KB of Claude Code's classifier prompt never gets printed
- A video editor with no timeline, driven from Claude Code
- Claude Fable 5's biology refusals drop about 85%
- Claude Code sessions can now run on your own machines
- Claude Fable now requires usage credits
- Project Panama: Anthropic's internal memo on destroying books after scanning
- Claude Fable 5 helped produce a counterexample to the Jacobian conjecture
- AI agents went loose on the live internet during UK safety tests
- Claude Code 2.1.223 patches a Bash permission bypass
- tcrf.net is serving Claude agents a prompt injection that wipes files
- Meta shipped three benchmark charts with Muse Code. Claude Opus 5 wins all three.
- Anthropic Is Building Its Own Chip
- Show HN: CostClaw, a local CLI that finds your Claude Code spend leaks
- The same three Claude models are throwing errors again
- Anthropic hires former California Supreme Court justice Tino Cuéllar
- Worktree isolation was leaking git commands into the main copy
- UK AI Security Institute: agents went after real code in testing
- Anthropic has signed a $10B deal for computing capacity with Volta
- Claude Code's deny rules don't stop grep
- Claude reviewing Codex: 71.6% to 89.7%. Codex reviewing Claude: 91.4% down to 82.8%.
- The new CLAUDE.md advice: delete, don't accumulate
- Claude Opus 4.5 squeezed a kernel from 147,734 cycles to 1,487
- Claude Code 2.1.221: permission bypass fixed, Focus view in VSCode, credential masking
- Amodei worries people are joining Anthropic for the paycheck
- A Claude session has been rewriting an app in Swift for over two weeks
- anthropickit harvested SSH keys during pip install
- Opus 5 falls to prompt injection 2% of the time
- An Amazon AI project blew through its budget by 860%
- Class action says Anthropic's Max 5x plan is closer to 3.5x