Claude Suffers a Partial Outage Across Claude.ai, Claude Code and Cowork
Anthropic confirmed “elevated error rates” around 10:21 AM ET on September 29. Users saw “Request couldn’t be processed” errors plus login and sign-up failures on Claude.ai, the desktop and mobile apps, Claude Code and Cowork.
Down Detector logged more than 11,800 reports by mid-morning. No cause or repair timeline was published as of the last update, only advice that retrying may succeed.
Anthropic: Open-Weight GLM-5.3 Brings Mythos-Class Cyber Skills to Everyone
Anthropic researchers say Zhipu AI’s freely downloadable GLM-5.3 can build end-to-end exploits comparable to Claude Mythos Preview. Its safeguards fell to simple tricks like deceptive prompts, prefilled reasoning and abliteration 64-100% of the time.
Human experts using it found previously unknown vulnerabilities in popular browsers. Anthropic’s point: advanced cyber capability is now broadly available without the access limits Mythos ships with.
“What Do You Want From AI?” Anthropic Opens a Public Survey
Anthropic launched a public engagement study using its Anthropic Interviewer tool, building on a December 2025 effort that reached 81,000 participants. Its stated view: how to weigh AI’s benefits and risks shouldn’t be left to AI companies alone.
Claude Code 2.1.285 Ships Desktop and Plugin Controls
The September 29 release adds desktop and plugin controls, improvements to the VS Code and web experiences, stronger model workflows, and a long list of fixes across sessions and permissions.
It follows 2.1.283 and 2.1.282 from September 25, which added gateway controls, MCP improvements and faster startup.
Plugin Directory Gives Developers a Self-Serve Path to Launch
The submission portal, live since September 25, offers auto-validation, review tracking and post-launch analytics. It supports MCP 2.0, MCP Apps and Enterprise Managed Auth.
It’s the on-ramp behind the Claude Marketplace we covered Monday.
Outage Day Puts Claude-Dependent Workflows in the Spotlight
With Claude Code and Cowork both hit, developers and teams leaning on Claude for daily work felt it fast. Retrying worked for some, but there was no ETA for a full fix.
Worth having a fallback model or queued-work plan for the next incident.
Sonnet 5.5 Is Now the Default Conversation in Dev Circles
Sonnet 5.5, released September 28, is live on the Claude API, Bedrock, Google Cloud and Microsoft Foundry. Releasebot flags breaking changes around thinking behavior and tool use that need code migrations before you switch.
Capability Is Spreading Faster Than Guardrails
Two stories, one theme. Reliability: when Claude wobbles, a lot of real work stops. Security: when a strong open-weight model ships without meaningful safeguards, the gating that closed models rely on stops being the whole story.
Anthropic’s survey is the softer counterpart: a bid to bring the public into the trade-offs. Expect more of this framing as the IPO nears.