Anthropic

Latest AI news, models and releases from Anthropic. ['anthropic-ai', 'Anthropic AI Models', 'Claude', 'Claude 2', 'Claude 2.1', 'Claude 3', 'Claude 3.5 Sonnet', 'Claude-3.5-Sonnet', 'Claude 3.5 Sonnet v2', 'Claude 3.7 Sonnet', 'Claude 3 Haiku', 'Claude 3 Opus', 'Claude3 Opus', 'Claude 3 Sonnet', 'Claude 4', 'Claude 4.6', 'Claude 4 Opus', 'Claude 4 Sonnet', 'Claude 5', 'Claude Agent SDK', 'Claude.ai', 'Claude AI', 'ClaudeBot', 'Claude Code', 'Claude Code (Opus 4.7)', 'Claude Code Sonnet 5', 'Claude Cowork', 'Claude Design', 'Claude Desktop', 'Claude Fable', 'Claude Fable 5', 'Claude Fable5', 'Claude Fable 5 Max', 'Claude Haiku', 'Claude Haiku 4.5', 'Claude Instant', 'Claude (language model)', 'Claude Max', 'Claude Mythos', 'Claude Mythos 5']

Official site ↗

AI Safety 🇩🇪

Few AI-Discovered Vulnerabilities Actually Exploited, Says VulnCheck Analysis

VulnCheck analyzed how often AI-discovered security vulnerabilities are actually exploited. Of 1,061 such vulnerabilities in H1 2026, only 14 (1.3%) had confirmed attacks, a rate similar to all vulnerabilities. However, exploitation is faster: average time from disclosure to first attack dropped to 80 days from 120.

Anthropic

The Decoder (DE)02.08 · 14:01

Agents 🇩🇪

Meta AI proposes a second AI agent to keep long tasks on track

Meta AI researchers have developed a memory module that uses a separate 'memory agent' to selectively inject reminders into an 'action agent' during long tasks, preventing 'behavioral state decay'. Tests on Terminal-Bench 2.0 and τ2-Bench showed significant improvements, especially for weaker agents, and the code is open-sourced on GitHub.