US EditionAnthropic AI Model Deemed Too Dangerous for Public Reportedly Accessed by Hackers
Anthropic said the model can find thousands of flaws in hundreds of systems, prompting a controlled response and fresh warnings from security leaders.
- A Discord group gained unauthorized access to Anthropic's Mythos Preview AI model on its debut day by guessing the model's online location using data from a recent breach at Mercor, an AI training startup.
- Using permissions from an Anthropic contracting firm, the individuals probed the system and "made an educated guess about the model's online location based on knowledge about the format Anthropic has used for other models," according to Bloomberg.
- Beyond Mythos, the group accessed other unreleased Anthropic models but reportedly limited their activity to building simple websites to avoid detection rather than conducting malicious hacking.
- The Cloud Security Alliance warned that AI accelerates vulnerability discovery, creating a "perfect storm" for defenders, while this incident challenges guardrails protecting powerful AI cybersecurity tools.
- Project Glasswing serves as Anthropic's controlled effort to use Mythos to secure critical software before comparable models become widely available, as the industry scrambles to defend against sophisticated nation-state and ransomware attackers.
12 Articles
12 Articles
Anthropic AI model deemed too dangerous for public reportedly accessed by hackers
SAN FRANCISCO (KRON) -- An artificial intelligence model that's been deemed too dangerous to be released to the public has reportedly been accessed by unauthorized users. The model, Claude Mythos, was announced by AI company Anthropic earlier this month. Mythos is a cybersecurity AI model that's apparently capable of finding otherwise undetectable software vulnerabilities in software systems with a minimum of human guidance. The model is said to…
Discord Users Breach Anthropic's Mythos AI in Security Incident #AI - National Cyber Security Consulting
The Buzz ■ Discord users gained unauthorized access to Anthropic's Mythos AI system, according to Wired ■ Spy firms exploited global telecom infrastructure weaknesses to track surveillance targets ■ 500,000 UK health records went up for sale on Alibaba's platform in a massive data breach ■ Apple patched a notification bug that could reveal sensitive […] Thank you for subscribing to our RSS feed! The post Discord Users Breach Anthropic's Mythos A…
Read the original note in the following link: Unauthorized access to Claude Mythos opens doubts about the security of AI models for cybersecurity International media detailed that a small group of unauthorized people would have accessed Claude Mythos Preview, an Anthropic model oriented to the detection of vulnerabilities in software. Anthropic indicated that it investigates the complaint and that, for now, it has no indication of access to the …
Coverage Details
Bias Distribution
- 50% of the sources lean Left, 50% of the sources are Center
Factuality
To view factuality data please Upgrade to Premium
