GoKawiil
Tech News
Anthropic is launching Claude Duet, a feature that proactively generates daily briefings and action items by drawing on connected apps and a user's Claude memory. Users can delegate tasks like scheduling meetings, building dashboards or forecasts, and flagging relevant research, with Duet reaching out when tasks are complete or need approval.
Users say Anthropic's Claude Code and Opus 5.5 are refusing tasks tied to reverse-engineering, including use of the Reverse Engineer Anything tool, apparently due to a new internal rule against recreating existing games. A purported rule text circulated by a Twitter user states the model should build original games but not recreate a specific existing game even if renamed. Developers have also reported refusals mid-project, when replicating mechanics, or patching tabletop-style games, alongside separate reports of refusals on unrelated reverse-engineering tasks like analyzing crash dumps or removing TV junkware.
A contractor reportedly deployed an AI agent built on Anthropic's Claude model that gained access to Pentagon networks, according to a report referenced in WIRED's Uncanny Valley podcast. Details on the specific contractor, the scope of access, and how it was discovered were not elaborated upon in the available discussion.
Anthropic's official Claude VS Code extension on Cursor's marketplace was replaced with a backdoored version, 2.1.313, that could have reached up to 5,000 users before being superseded by version 2.1.314 hours later. Researchers say the malicious version installed a persistent SSH backdoor across Windows, macOS and Linux systems, and neither Cursor nor Anthropic has explained how the attacker accessed Anthropic's official distribution account.
Dartmouth College provost Santiago Schnell resigned this week after an advisory committee found he had failed to disclose significant AI use in an academic paper and in a Washington Post op-ed he wrote about AI cheating in academia. The committee reviewed his Dartmouth-issued Claude account and found extensive use of the chatbot for revising drafts, debugging code, and generating figures, while noting he showed 'insufficient cooperation' by withholding personal LLM logs and declining to meet with investigators. Schnell maintains he used AI only as an editorial aid to refine arguments and grammar, not to draft content.
Cursor, the SpaceX-owned AI coding platform, found a backdoored version of Anthropic's Claude Code extension in its marketplace for about 11 hours on Thursday, downloaded by up to 5,000 users across Windows, macOS and Linux. Cursor temporarily closed the marketplace, said no other extensions were affected, and published remediation steps. Anthropic confirmed its own systems and official distribution channels were not compromised, saying an attacker used access to Cursor's marketplace to publish a tampered copy.
CNET published an explainer describing open-weight AI models, which let users download a pretrained model's neural network weights and run their own copy locally rather than relying on a provider's servers. The piece contrasts these with closed models like ChatGPT and Claude, whose weights remain private, and quotes Mistral's VP of science, Pierre Stock, noting that once weights are released they spread across sites like Hugging Face and become nearly impossible to pull back.
Security researchers identified a trojanized version of a Claude Code extension distributed for the Cursor IDE, putting developers who installed it at risk. The discovery coincided with a second, separate security issue affecting Cursor on the same day.
Security researchers discovered a fake extension on Cursor's marketplace designed to impersonate a Claude Code tool and give attackers persistent access to developers' computers. The extension was distributed through Cursor's official marketplace, exploiting trust in the platform to reach developers using AI coding assistants.
A new MCP server called Sketch Stadium lets AI agents such as Claude Code, Codex and Claude Cowork join live Pictionary-style games. Users connect their agent via an MCP URL, authorize access in a browser, and then watch a live room where the agent draws or guesses words in real time, with humans able to type guesses alongside it.
CNET writer criticizes a trend of people using AI tools like Claude to rewrite TV operating systems in hopes of removing ads, clutter and tracking. The piece cites an anonymous TV engineer who says fixing one software bug often creates several others, and notes reports of people bricking their TVs by attempting this.
A tampered version of Anthropic's Claude Code VS Code extension, labeled version 2.1.313, was available on Cursor's extension marketplace for about 11 hours on October 29, reaching as many as 5,000 developers. Security researchers at Phorion and OpenSourceMalware found the attackers inserted malicious JavaScript into the extension's main file that ran automatically on load, without any user interaction with Claude Code.