claude anthropic ai-safety cybersecurity

New serious vulnerabilities spiked around release of Claude Mythos Preview

Disclosed CVEs: 3.5Ã Spike After Claude Mythos

epoch.ai

July 3, 2026

1 min read

🔥🔥🔥🔥🔥

55/100

Summary

In April 2026, Anthropic introduced Claude Mythos Preview, an internal model designed for autonomous cybersecurity vulnerability discovery and exploitation. Following this announcement, both Anthropic and OpenAI initiated projects to enhance critical software security against potential threats.

Key Takeaways

In April 2026, Anthropic announced the Claude Mythos Preview model, which can autonomously discover and exploit cybersecurity vulnerabilities.
Following the announcement of Claude Mythos, the number of high- and critical-severity Common Vulnerabilities and Exposures (CVEs) increased by more than 3.5 times in June compared to the previous monthly record.
Both Anthropic and OpenAI are working to use advanced models to strengthen critical software against potential exploitation by malicious actors.

Read original article

View story →

Coverage & discussions

Hacker News · epoch.ai

New serious vulnerabilities spiked around release of Claude Mythos Preview ↗

Community discussions

Hacker News discussion ↗

Historical count: 5 · 5 saved in sample · Historical sample

Analysis ready

Community Sentiment

Negative

Positives

Project Glasswing users are taking a rigorous approach to verifying vulnerabilities, ensuring that the reported issues with Mythos are credible and not just hallucinations.
The discussion around AI-assisted vulnerabilities highlights the proactive stance some security engineers are taking, emphasizing the importance of careful validation over blind reporting.

Concerns

The spike in vulnerabilities suggests that either bad actors are exploiting weaknesses or that the quality of AI models like Mythos is degrading, leaving systems open to sophisticated exploits.
There's skepticism about the validity of the vulnerability reports, with concerns that many could be hallucinations rather than real issues stemming from the AI's claims.
The rush to fix vulnerabilities based on AI-generated reports could lead to hasty decisions, giving Anthropic disproportionate power over the security landscape.