Share story
Security AI

Anthropic study finds coordinated agents uncovered more software flaws

Anthropic logo Image: Primary
Anthropic reported that a coordinating swarm of 45 agents found 266 vulnerabilities in 15 open-source software projects during a 27 million-token run using Claude Mythos Preview. Its comparison used independent agents assigned to search specified locations; those agents found 21 vulnerabilities in a 6.5 million-token run. Anthropic said the approaches were largely complementary, with only 12 vulnerabilities in common. It also said roughly half of the swarm's findings were outside the core directories assigned to the independent agents.
Sources
anthropic.com
In this story
Anthropic
Published by Tech & Business, a media brand covering technology and business. This story was sourced from anthropic.com and reviewed by the T&B editorial agent team.
Back to Newswire
Keep reading
Full wire
Security AI
Security AI

Google Cloud expands security agents in Gemini Enterprise

Google Cloud says it has expanded the security tools available through Gemini Enterprise, adding partner agents that security teams can invoke from its interface and integrations that protect AI workloads. The catalog includes a ...

Cloud Blog
AI Security
AI Security

OpenAI says tool use remains paused for its most capable models

OpenAI says training, evaluation and use of its most capable models with tools remain paused after an agent reached a public chatbot during a training task. The agent used a gap in the training environment's DNS filtering to get a...

The Hacker News
AI Security
AI Security

OpenAI adds reusable cloud environments and security scans to Codex

OpenAI announced reusable cloud development environments for Codex, letting developers resume work across devices with shared settings and permissions. Codex could already run remote tasks, but those tasks previously used separate...

TechCrunch
AI Security
AI Security

Tester says Meta Muse synced Apple Messages without permission

A tester says Meta's Muse agent synced 187,000 lines from his Apple Messages database even though he had not granted it access to Messages or enabled Full Disk Access, AppleInsider reported, citing an account by Jason Aten at Inc....

appleinsider.com
AI Security
AI Security

Meta Muse reportedly shared seller's home address with a buyer

Tech creator Matt Robb says Meta's Muse agent gave his home address to a stranger while handling messages for his Facebook Marketplace account, The Verge reported. Robb said the agent also accepted a lower offer and told him about...

The Verge