Anthropic, Claude & AI Models
Anthropic released Claude Opus 4.7, the latest version of their flagship model. The HN discussion covers benchmarks, capabilities, and comparisons to previous versions.
Simon Willison analyzes the differences between Claude Opus 4.6 and 4.7 system prompts, noting that Anthropic remains the only major AI lab to publicly publish the system prompts for their user-facing chat systems.
A critical analysis of Anthropic's claims around the Claude Mythos model launch, arguing that some of the marketing and benchmarks involve misleading information.
Financial Times · www.ft.com
Anthropic's new Mythos system has sparked fears it could turbocharge hacking and expose cybersecurity weaknesses faster than they can be patched.
A leaked version of Anthropic's Claude Code tool revealed critical command injection vulnerabilities, raising concerns about security in AI-powered coding tools.
Privacy & Security
A privacy researcher discovered that Claude Desktop silently installs undocumented browser extensions in Chrome and other browsers, raising concerns about user consent and data collection.
A Lawfare article argues it is time to ban the commercial sale of precise geolocation data, citing the serious privacy and national security risks posed by the data broker industry.
A security researcher discovered that Notion exposes the email addresses of all editors on any public page, creating a privacy leak affecting potentially millions of users.
Vercel confirmed a security breach in which hackers claim to be selling stolen data, prompting the company to publish an incident report and notify affected users.