Substrate
Topic

software-vulnerabilities

6 stories related to this topic, newest first.

Anthropic Reports Over 10,000 High-Severity Bugs Found in Project Glasswingnbcnews.com
technology6 days agoDeveloping

Anthropic Reports Over 10,000 High-Severity Bugs Found in Project Glasswing

Anthropic said its AI-assisted security testing program has identified more than 10,000 high- or critical-severity vulnerabilities. The company is working with about 50 partner organizations and expanding the project to additional participants.

Benzinga
1 source
Startup Depthfirst AI Model Finds Vulnerabilities Missed by Anthropic Mythosnbcnews.com
ai7 days ago

Startup Depthfirst AI Model Finds Vulnerabilities Missed by Anthropic Mythos

Depthfirst reported that its AI model identified additional security flaws in widely used internet software. The company said the findings cost one-tenth the price of comparable results from Anthropic's Mythos model.

FO
1 source
Microsoft MDASH Outperforms Mythos Preview on CyberGym Benchmarkforbes.com
technology14 days agoDeveloping

Microsoft MDASH Outperforms Mythos Preview on CyberGym Benchmark

Microsoft announced MDASH, a multi-modal agentic scanning harness for vulnerability discovery and remediation. The system scored 88.4 percent on the CyberGym benchmark compared with 83.1 percent for Mythos Preview. MDASH uses more than 100 specialized agents and is in limited pri…

forbes.com
1 source
Researcher Reports Two Flaws in Gmail and Google Drive IntegrationForbes
technology17 days agoDeveloping

Researcher Reports Two Flaws in Gmail and Google Drive Integration

Security researcher Ben Ilkashi identified an architectural misalignment that allows malicious files blocked by Gmail to be delivered via Google Drive with a "Scanned by Gmail" label. A second flaw permits bypassing of Google Drive's incomplete scan warnings when files are access…

Forbes
1 source
Google Reports Hackers Used A.I. to Discover Software Flaw680news.com
ai18 days ago

Google Reports Hackers Used A.I. to Discover Software Flaw

Google said a criminal hacking group attempted a widespread cyberattack that leveraged artificial intelligence to identify a previously unknown vulnerability. The company reported it had high confidence the actors used an A.I. model to discover and weaponize the flaw. The inciden…

The New York Times
680news.com
flipboard.com
winnipegfreepress.com
4 sources
Anthropic's Mythos AI Revealed for Hacking Capabilities and Limited AccessSubstrate placeholder — needs review
ai35 days agoDeveloping

Anthropic's Mythos AI Revealed for Hacking Capabilities and Limited Access

Anthropic developed an AI model called Mythos capable of identifying software vulnerabilities quickly. The company kept it private due to security concerns but is sharing it with select firms under Project Glasswing. Experts note it aids in finding flaws but does not uncover uniq…

New Scientist
1 source