{ "no": 10, "generated_at": "2026-08-15T06:50:00-04:00", "entries": [ { "date": "2026-08-15", "seed": "2026-08-15", "lede": { "title": "Anthropic explains how Claude's new text watermark works", "summary": "Anthropic published a detailed explainer on the invisible text watermarking it has begun applying to Claude's output, built on Google DeepMind's SynthID-Text technique, which subtly biases word choice among equally valid options using a cryptographic key rather than adding visible markers. The rollout — which doesn't affect quality, speed, or cost, and can't identify individual users — is Anthropic's response to the EU AI Act's Code of Practice on transparency for AI-generated content, signed by roughly 190 organizations. It applies globally to future Claude models by default; watermarking for models launched before August 2, 2026 is being added retroactively over the coming months, with a detection API coming soon.", "url": "https://www.anthropic.com/news/claude-text-watermark", "sources": [ "https://www.anthropic.com/news/claude-text-watermark", "https://techcrunch.com/2026/08/11/anthropic-says-it-will-watermark-text-generated-by-its-ai-models/", "https://www.euronews.com/next/2026/08/11/eu-compliance-delivered-globally-anthropic-to-watermark-claudes-output-worldwide" ] }, "briefs": [ { "tag": "Anthropic Research · Aug 14", "title": "Risk report discloses 'Model 2,' an internal model more capable than Mythos 5", "summary": "Anthropic's latest Risk Report raised its assessment of misalignment risk in high-stakes settings from \"very low\" to \"low\" — a change the company says reflects increased uncertainty from recent cybersecurity-evaluation incidents, not new evidence of danger. The same report disclosed \"Model 2,\" an unreleased internal model \"somewhat more capable\" than Mythos 5 that Anthropic staff already use for engineering work; the company says it has no current plans to release it externally and hasn't finished its full predeployment safety suite on it.", "url": "https://www.anthropic.com/aug-2026-risk-report", "sources": [ "https://www.anthropic.com/aug-2026-risk-report", "https://siliconangle.com/2026/08/14/anthropic-details-unreleased-model-2-new-alignment-concerns-latest-ai-risk-report/" ] }, { "tag": "Claude Code · Aug 14", "title": "Claude Code 2.1.233 pulls task-tracking tools from newer models by default", "summary": "Version 2.1.233, released Aug 14, turns off TaskCreate/Get/Update/List and TodoWrite by default on Opus 4.8, Sonnet 5, Fable 5, Mythos 5, and newer models (set CLAUDE_CODE_ENABLE_TODO_TOOLS=1 to restore them), adds GitLab merge-request support to --worktree and the agents view, and fixes a Windows regression where auto mode kept stopping for approval on ordinary cd && command > file lines.", "url": "https://code.claude.com/docs/en/changelog", "sources": [ "https://code.claude.com/docs/en/changelog" ] } ] }, { "date": "2026-08-14", "seed": "2026-08-14", "lede": { "title": "Claude Code makes auto mode the default today for Pro, Max, and Team", "summary": "Starting today, Aug 14, Claude Code's auto mode — a classifier that screens actions instead of prompting for every edit or command — becomes the default permission mode for Pro, Max, and Team plans; Enterprise, API, and cloud-platform versions stay opt-in for now, with Anthropic planning to follow within the month. Anthropic says internal red-teaming, third-party evals, and a 1,053-user study found the classifier caught 89% of deliberately inserted dangerous commands versus 13.6% for human reviewers, who approve 97% of permission prompts they see. Auto mode falls back to manual approval after three consecutive blocks or 20 in a session, and the switch ships alongside new prompt-injection screening and customizable hard-deny rules.", "url": "https://code.claude.com/docs/en/permission-modes", "sources": [ "https://code.claude.com/docs/en/permission-modes", "https://techcrunch.com/2026/08/09/anthropic-is-turning-claude-codes-auto-mode-on-by-default/", "https://www.theregister.com/ai-and-ml/2026/08/10/claude-code-puts-auto-mode-in-the-drivers-seat/5285326" ] }, "briefs": [ { "tag": "Claude Developer Platform · Aug 11", "title": "Compliance API now covers local Cowork and Claude Code session transcripts", "summary": "New beta endpoints for Claude Enterprise let admins pull transcripts of Cowork and Claude Code sessions that run on users' own machines — list sessions org-wide, or fetch one session's metadata or full transcript — with the existing Compliance Access Key. The same release added an anthropic-workspace-id response header identifying which workspace resolved a request.", "url": "https://platform.claude.com/docs/en/release-notes/overview", "sources": [ "https://platform.claude.com/docs/en/release-notes/overview" ] }, { "tag": "Claude Developer Platform · Aug 10", "title": "Sonnet 5's launch pricing becomes the standard rate", "summary": "The $2/$10-per-MTok introductory price for Claude Sonnet 5 is now the standard rate — the previously scheduled increase to $3/$15 on September 1 will not occur.", "url": "https://platform.claude.com/docs/en/release-notes/overview", "sources": [ "https://platform.claude.com/docs/en/release-notes/overview" ] } ] }, { "date": "2026-08-13", "seed": "2026-08-13", "lede": { "title": "Claude Code 2.1.229 ships plugin marketplace command sources and streaming fixes", "summary": "Version 2.1.229, released Aug 12, adds plugin marketplace \"command\" sources — a local command (like an IDE) can print a plugin directory that's re-resolved each session and applied without a restart — plus server-supplied Claude Code hook support for self-hosted runner sessions and SSE keepalive pings on gateway streaming to stop idle-timeout disconnects on Vertex and Bedrock. ListAgents now marks disconnected Remote Control sessions as offline and labels cloud sessions as cloud. The 32-change release also fixes long responses partly disappearing and printing twice while streaming, a RangeError crash when rendering in very narrow terminals, MCP OAuth failing with strict authorization servers by switching to 127.0.0.1 instead of localhost, and the Claude Code Review GitHub workflow completing without posting its review.", "url": "https://code.claude.com/docs/en/changelog", "sources": [ "https://code.claude.com/docs/en/changelog" ] }, "briefs": [ { "tag": "Claude Code · Aug 13", "title": "2.1.231 fixes MCP OAuth sign-in for pre-registered clients like Slack", "summary": "A single-fix point release: 2.1.231 patches MCP OAuth sign-in failing with a redirect URI mismatch for servers that use a pre-registered OAuth client, such as Slack's MCP server.", "url": "https://code.claude.com/docs/en/changelog", "sources": [ "https://code.claude.com/docs/en/changelog" ] } ] }, { "date": "2026-08-12", "seed": "2026-08-12", "lede": { "title": "Claude Code 2.1.228 hardens synced skills against command shadowing", "summary": "Version 2.1.228, released Aug 11, hardens skills synced from claude.ai so they can no longer shadow local commands or MCP prompts; their descriptions are now sanitized and labeled, and on your machine their bodies can't run `!` shell commands or expand `@` files. The same 18-change release fixes interactive sessions that could freeze mid-redraw after a rare layout error, git/Git Bash not being found on Windows when Claude Code launches from a parent folder of the git install, `/tui` reverting to an earlier model after a `/model` change, and cross-session messaging sometimes starting without an inbox after install or upgrade — plus it lets newer models overwrite a file they haven't read this session, matching the Edit tool's rules, and removes an outdated notice about auto mode costing more.", "url": "https://code.claude.com/docs/en/changelog", "sources": [ "https://code.claude.com/docs/en/changelog", "https://github.com/anthropics/claude-code/releases/tag/v2.1.228" ] }, "briefs": [ { "tag": "Claude Developer Platform · Aug 11", "title": "Compliance API now covers Cowork and Claude Code sessions", "summary": "In beta for Claude Enterprise, the Compliance API now extends beyond claude.ai chat to Cowork (desktop, web, mobile) and Claude Code (CLI, desktop app) sessions running on users' own machines. New endpoints return consolidated transcripts — prompts, responses, tool activity, timestamps — via the existing Compliance Access Key; current API users need no separate integration. The beta excludes Claude Code on web, via the Claude Platform, and sessions on Bedrock, Vertex AI, or Microsoft Foundry.", "url": "https://claude.com/blog/compliance-api-cowork-and-claude-code", "sources": [ "https://claude.com/blog/compliance-api-cowork-and-claude-code", "https://platform.claude.com/docs/en/release-notes/overview" ] } ] }, { "date": "2026-08-11", "seed": "2026-08-11", "lede": { "title": "Claude Code 2.1.227 fixes a login-token bug that misfired Fable credit prompts", "summary": "Version 2.1.227, released Aug 10, patches a bug where a session that started with an expired login token evaluated feature flags without checking the user's subscription tier — wrongly nudging Max plan users to enable usage credits for Claude Fable. The same release fixes claude-code-action Bash commands failing outright under allowed_non_write_users on GitHub-hosted runners, fixes /tui restoring a conversation that had been rewound to before its first message, and polishes the slash-command menu (cleaner selection highlighting, bolded matched characters, correct glyphs for emoji or accented command names) plus event-loop performance around file-not-found suggestions. It's a maintenance release rather than a headline feature — Anthropic's own release notes and newsroom have stayed quiet since Aug 7-8, so this patch, dated yesterday, is the freshest verified item across every official source checked this morning.", "url": "https://code.claude.com/docs/en/changelog", "sources": [ "https://code.claude.com/docs/en/changelog", "https://raw.githubusercontent.com/anthropics/claude-code/main/CHANGELOG.md" ] }, "briefs": [] }, { "date": "2026-08-10", "seed": "2026-08-10", "lede": { "title": "Claude Code's auto mode becomes the default for Pro, Max, and Team on Aug 14", "summary": "Starting August 14, Claude Code will run in auto mode by default for Pro, Max, and Team plans, replacing repeated permission prompts with a classifier that flags irreversible, destructive, or out-of-bounds actions instead. Anthropic says testing with 1,053 paid users found the classifier caught 89% of dangerous commands versus 13.6% for manual human review. Claude Enterprise, the Claude API, Claude Platform on AWS, Amazon Bedrock, Google Cloud's Agent Platform, and Microsoft Foundry get the new default within the coming month. Boris Cherny, Head of Claude Code, said in a post on X, per TechCrunch: \"The team and I use Auto mode exclusively, and have been for many months. I couldn't imagine going back to permission prompts!\" It's a quiet-news stretch otherwise — official channels have gone dark since Aug 7-8, so this rollout, still generating fresh coverage today, is the freshest live story worth knowing.", "url": "https://claude.com/blog/auto-mode-default-in-claude-code", "sources": [ "https://claude.com/blog/auto-mode-default-in-claude-code", "https://techcrunch.com/2026/08/09/anthropic-is-turning-claude-codes-auto-mode-on-by-default/" ] }, "briefs": [ { "tag": "Claude Code · Security · Aug 7", "title": "Third-party test: Claude Code's auto mode blocked all 720 prompt-injection attempts", "summary": "Independent red-teaming by Trajectory Labs ran 72 indirect prompt-injection scenarios, each attempted 10 times (720 attempts total), against Claude Fable 5, Opus 5, and Sonnet 5 running Claude Code's auto mode — none succeeded. A comparable agent running GPT-5.6 Sol in its auto-review mode had a 5.83% success rate against the same attacks. Anthropic cited the results, plus new prompt-injection screening and customizable \"hard deny\" rules, in its case for making auto mode the default.", "url": "https://claude.com/blog/auto-mode-default-in-claude-code", "sources": [ "https://claude.com/blog/auto-mode-default-in-claude-code" ] } ] }, { "date": "2026-08-09", "seed": "2026-08-09", "lede": { "title": "Anthropic loosens Fable 5's biology safeguards, cutting false-positive fallbacks by 85%", "summary": "On Aug 7, Anthropic said it rewrote Fable 5's biology-safety classifiers with expert review to cut false-positive \"fallbacks\" — cases where a benign question gets rerouted to the less-capable Opus 5 model — by about 85% overall, testing found. Anthropic expects the steepest drops on claude.ai (67%) and Cowork (55%), with smaller gains on Claude Code (17%) and the Claude Platform (7%). The model still blocks dual-use queries touching virology, toxicology, and molecular design pending \"trusted access pathways,\" but clinicians, students, and researchers should see far fewer benign questions get punted. It's been the most substantive Anthropic news in days — Aug 8 brought two Claude Code releases (one substantial, one routine) and Aug 9 has been quiet so far.", "url": "https://www.anthropic.com/news/improving-fable-5-s-biology-safeguards", "sources": [ "https://www.anthropic.com/news/improving-fable-5-s-biology-safeguards", "https://thenextweb.com/news/anthropic-claude-fable-5-biology-safeguards-fallbacks-dual-use" ] }, "briefs": [ { "tag": "Claude Code · v2.1.226 · Aug 8", "title": "Claude Code ships a quiet reliability patch", "summary": "Version 2.1.226, released Aug 8, is described in the official changelog simply as \"bug fixes and reliability improvements\" — no user-facing feature changes are called out, a quiet follow-on to earlier the same day's more substantial 2.1.225 release.", "url": "https://code.claude.com/docs/en/changelog", "sources": [ "https://code.claude.com/docs/en/changelog" ] }, { "tag": "Claude Apps · Aug 6", "title": "Enterprise plans get beta security scanning for skills and plugins", "summary": "Claude Enterprise customers can now turn on a beta feature that automatically scans third-party skills and plugins for malicious content whenever someone uploads or edits them, adding a supply-chain safeguard as Skills and plugin marketplaces expand.", "url": "https://support.claude.com/en/articles/12138966-release-notes", "sources": [ "https://support.claude.com/en/articles/12138966-release-notes" ] } ] }, { "date": "2026-08-08", "seed": "2026-08-08", "lede": { "title": "Claude Managed Agents gain session budgets, an advisor mode, and data-residency controls", "summary": "On Aug 7, Anthropic added four capabilities to Claude Managed Agents on the Developer Platform: session budgets that hard-cap spend at public list rates and pause a session with a budget_reached stop reason when hit; an \"advisor\" role letting the primary agent consult a separate, at-least-as-capable model mid-turn for strategic guidance; per-agent data residency control via inference_geo, letting teams choose where inference runs; and automatic loading of skills from a session's mounted GitHub repository. Together the updates give teams building long-running agents on the platform much finer control over cost, geography, and capability.", "url": "https://platform.claude.com/docs/en/release-notes/overview", "sources": [ "https://platform.claude.com/docs/en/release-notes/overview", "https://releasebot.io/updates/anthropic/claude-developer-platform" ] }, "briefs": [ { "tag": "Claude Code · v2.1.225 · Aug 8", "title": "Claude Code adds gateway spend-limit warnings and a workspace-trust prompt for agents", "summary": "Today's release names the spend cap, reset time, and operator message when a gateway limit is hit, adds a workspace-trust prompt to claude agents for untrusted directories, and fixes a clutch of reliability bugs — a transient 401 wrongly swapping a long-lived OAuth token for a short-lived one (breaking headless sessions until restart), a macOS keychain timeout causing spurious MCP 401 errors, and cross-session messages getting stuck without expiry in headless runs.", "url": "https://code.claude.com/docs/en/changelog", "sources": [ "https://code.claude.com/docs/en/changelog" ] }, { "tag": "Claude Code · v2.1.224 · Aug 7", "title": "Claude Code lets teams run sessions on their own compute, message across machines", "summary": "The Aug 7 release adds claude self-hosted-runner, letting Team/Enterprise admins point Claude Code's web, mobile, and desktop sessions at their own machines or containers; introduces cross-session SendMessage (macOS/Linux) so sessions on different machines can message each other via ListAgents; and adds a zip-based archive plugin source with optional SHA-256 pinning plus JWT-aware sandbox credential masking.", "url": "https://code.claude.com/docs/en/changelog", "sources": [ "https://code.claude.com/docs/en/changelog" ] } ] }, { "date": "2026-08-07", "seed": "2026-08-07", "lede": { "title": "Anthropic retunes Claude Fable 5's biology safeguards, cutting biology-related fallbacks 85%", "summary": "Anthropic narrowed Fable 5's biology-safety classifier on Aug 7 so it stops blocking benign questions — interpreting lab results, explaining symptoms, general biology homework — while continuing to fall back to Opus 5 on dual-use requests around virology, toxicology, and molecular design. In testing, biology-related fallbacks dropped by about 85%, with total fallback reductions of roughly 67% on Claude.ai, 55% on Cowork, 17% on Claude Code, and 7% on the Claude Platform. Anthropic says it consulted a range of internal and external experts before moving the classifier boundary.", "url": "https://www.anthropic.com/news/improving-fable-5-s-biology-safeguards", "sources": [ "https://www.anthropic.com/news/improving-fable-5-s-biology-safeguards", "https://www.unite.ai/anthropic-retunes-fable-5s-biology-safeguards-cutting-blocked-queries-85/", "https://thenextweb.com/news/anthropic-claude-fable-5-biology-safeguards-fallbacks-dual-use" ] }, "briefs": [ { "tag": "Leadership", "title": "Tino Cuéllar joins Anthropic as its first Chief Global Affairs Officer", "summary": "The former California Supreme Court justice and Carnegie Endowment president will lead Anthropic's global policy, diplomacy, and government-relations work — announced Aug 4 as its first person in the role.", "url": "https://www.anthropic.com/news/tino-cuellar", "sources": [ "https://www.anthropic.com/news/tino-cuellar", "https://www.thecrimson.com/article/2026/8/4/cuellar-anthropic-global-affairs/" ] }, { "tag": "Claude Code · v2.1.223", "title": "Claude Code adds marketplace org-wildcards and closes a Bash permission bypass", "summary": "The Aug 6 release lets admins allow or block whole GitHub orgs in marketplace settings via owner/* wildcards, adds a /teleport hint for continuing cloud sessions locally, folds /review into /code-review, and fixes a bug where tab- or invisible-Unicode-padded commands could hide part of themselves from permission prompts.", "url": "https://code.claude.com/docs/en/changelog", "sources": [ "https://code.claude.com/docs/en/changelog", "https://releasebot.io/updates/anthropic" ] }, { "tag": "Claude Developer Platform", "title": "Claude Enterprise gets inference hooks, an inline DLP checkpoint for every prompt", "summary": "Announced Aug 5, Enterprise customers can route prompts and tool responses through their own security server for an allow/deny check before Claude ever sees them — spanning Claude chat, Claude Code, and Cowork, with shadow mode, gradual rollout, and compatibility with tools like Netskope and Zscaler.", "url": "https://claude.com/blog/claude-enterprise-inference-hooks", "sources": [ "https://claude.com/blog/claude-enterprise-inference-hooks", "https://www.unite.ai/anthropic-puts-inline-data-loss-prevention-inside-claude-enterprise/" ] } ] }, { "date": "2026-08-07", "seed": "2026-08-07", "lede": { "title": "Anthropic ships Claude Opus 5, its new flagship for long-running agents", "summary": "Opus 5 landed July 24 as Anthropic's top-tier model, pitched as a step change for long-running agentic work and coding, with a cost/capability toggle.", "url": "https://www.anthropic.com/news/introducing-claude-opus-5", "sources": [ "https://www.anthropic.com/news/introducing-claude-opus-5", "https://techcrunch.com/2026/07/24/anthropic-launches-opus-5/", "https://fortune.com/2026/07/24/anthropic-debuts-claude-opus-5-with-feature-that-lets-users-toggle-between-cost-and-capability/" ] }, "briefs": [ { "tag": "Claude Code · v2.1.224", "title": "Self-hosted runners and cross-session messaging land in Claude Code", "summary": "Aug 7 release adds claude self-hosted-runner for Team/Enterprise, cross-session SendMessage via ListAgents, a zip-based plugin source with SHA-256 pinning, and JWT-aware sandbox credential masking.", "url": "https://code.claude.com/docs/en/changelog", "sources": [ "https://code.claude.com/docs/en/changelog", "https://raw.githubusercontent.com/anthropics/claude-code/main/CHANGELOG.md" ] }, { "tag": "Leadership", "title": "Tino Cuéllar joins as Anthropic's first Chief Global Affairs Officer", "summary": "The former California Supreme Court justice and Carnegie Endowment president will lead Anthropic's AI-policy and diplomacy strategy, announced Aug 4.", "url": "https://www.anthropic.com/news/tino-cuellar", "sources": [ "https://www.anthropic.com/news/tino-cuellar", "https://www.cnbc.com/2026/08/04/anthropic-names-global-affairs-chief-as-trump-tensions-persist.html" ] }, { "tag": "Safety", "title": "Anthropic retunes Fable 5's biology safeguards", "summary": "A same-day tightening-and-loosening pass on Claude Fable 5's bio-risk filters, aimed at cutting false-positive refusals while holding the line on genuine dual-use risk.", "url": "https://www.anthropic.com/news/improving-fable-5-s-biology-safeguards", "sources": [ "https://www.anthropic.com/news/improving-fable-5-s-biology-safeguards" ] }, { "tag": "Policy", "title": "Anthropic publishes its position on open-weights models", "summary": "A July 27 stance piece on how Anthropic thinks about open-weight releases versus its own closed-weight approach.", "url": "https://www.anthropic.com/news", "sources": [ "https://www.anthropic.com/news" ] } ] } ] }