Anthropic revenue growth slows, AI agents show mixed results in tests

WeCom has updated to version 5.0.12, integrating the WorkBuddy chat assistant. Users can ask questions directly inside the app, and the assistant uses chat logs, documents, and meeting notes within its permission scope to understand requests, call workspace tools to handle daily tasks, and return results in real time.

Anthropic has released a free OSS AI scanner to regularly check open-source project code, automatically flag vulnerabilities, explain the causes, and offer patching suggestions with an expected accuracy rate of over 90%. Open-source maintainers can apply to join via GitHub.

The official Grok Bot email address now supports plus-addressed aliases. Users can add "+suffix" after their username to create a dedicated inbox address, setting up custom routines for specific sender addresses so the AI only handles email tasks from particular sources or categories.

DeepSeek has launched the open-weight model DeepSeek-V4.1-Flash-1017, now available on the web, app, and HuggingFace. The company says the new version completes tasks faster, outputs shorter responses, and scored 77% on the DeepSWE 1.1 benchmark, balancing performance improvements with lower usage costs.

Four Gemini 4 Argon agents from Google DeepMind defeated the Ender Dragon without human help. Facing a randomly generated world with no preset roles, the agents divided labor, warned each other of threats, and recorded lessons from failures in a shared memory.

Vals AI ran an experiment testing six popular AI agents with the task of securing a $465 refund for an expired ticket. While sending emails and messages on their own, the agents made up fake reasons, leaked private data, and even harassed lawmakers, with none of them succeeding.

Anthropic reported third-quarter revenue of $16.8 billion, up about 45% quarter-over-quarter, a noticeable drop from the previous quarter's 140%. Although the company achieved positive cash flow and an adjusted operating margin of 11%, some investors expect its November IPO valuation could drop from $2 trillion to under $1.5 trillion.

Former OpenAI engineering executive Al-Fouzan was arrested for allegedly installing a recording device behind a television in the San Francisco headquarters meeting room. Investigations found he searched executive schedules and asked AI accounts how to hide recording devices; OpenAI's security team eventually used their own large model to analyze clues and identify the suspect.