Commenters focused on a malicious version of the Claude Code VS Code extension, with a security report describing an SSH key added to users’ authorized_keys. They disputed where the compromised version was distributed and who was responsible: Cursor said it was not developed by Anthropic, while commenters questioned whether Anthropic’s publishing account, Open VSX, or another part of the supply chain had been compromised. Some believed the affected release was limited to Open VSX and Cursor; others pointed to reports or screenshots involving Microsoft’s marketplace. They called for Anthropic to explain how the tampering happened, beyond providing instructions to remove it.
Many commenters treated the incident as a warning about trusting extensions and agentic coding tools with access to files and credentials. They criticized weak supply-chain defenses and automatic updates, and some recommended running agents in isolated virtual machines. Others argued that a compromised extension does not prove Claude Code or Codex is inherently unsafe, since any installed software can be attacked. Commenters also questioned claims that advanced AI tools make software development or security effectively solved. One commenter speculated that the extension attack might be connected to a Binance theft, while others focused on the risks to developer accounts and downstream software.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.
A Linux kernel use-after-free vulnerability, CVE-2026-98432, can enable remote code execution, and Go 1.24 enables MPTCP by default on Linux, exposing listeners to the demonstrated exploit. Upgrade to a patched kernel; disabling MPTCP in Go is only a fragile mitigation. (groups.google.com)
Scribd disclosed a security incident in which an unauthorized party may have accessed users’ email addresses, partial payment card details, country, issuer, expiry date, and payment history. The company says it secured the affected system, notified authorities, and advises users to watch for phishing. (news.ycombinator.com)
Flock camera searches in Highland Heights fell 95% after a dispatcher was caught misusing the system. The sharp drop followed the discovery of the misconduct. (wkyc.com)
AT Protocol apps can store user data publicly on personal data servers, where it is easy to search and retrieve. Skyreader subscriptions exposed more than 1,300 RSS feeds, including private and paywalled feeds, while other apps also publish messages openly. (melaniekat.com)
At least nine AI developer platforms disclosed breaches in October, with attackers targeting stored credentials and environment variables that grant access to AI services. The incidents point to coordinated agent-driven attacks fueled by the resale of stolen AI access, though investigations are still ongoing. (wiz.io)
The CVE-2026-112816 page requires JavaScript to display properly. The supplied content gives no details about the vulnerability. (cve.org)
Today's best Hacker News stories, summarized and screenshotted, one email a day.