Anthropic Limits AI Release Due to Risk
Anthropic is delaying the release of its new AI model, Mythos, because it is too powerful. The AI can find many security problems in computer systems. This raises concerns about how dangerous advanced AI can be.
Anthropic developed a new AI model called Mythos that can identify thousands of security vulnerabilities in software. Approximately fifty companies will initially have access to the model through the Project Glasswing program. The company is limiting access to only 40 security organizations due to concerns about the AI’s capabilities. Experts believe the AI is too dangerous for current cybersecurity standards. Anthropic is working with companies like Google, Apple, and Microsoft to test the AI and address vulnerabilities. This situation has led to concerns about potential misuse and prompted emergency measures from central banks and intelligence agencies.
Summarized from the sources above. Read the originals for the full story.
Highlights
AI Model Restricted for Security Risks
Anthropic is limiting the release of its AI model, Mythos, due to concerns about potential cyberattacks.
Model Identifies Many Vulnerabilities
The AI model, Mythos, has identified thousands of software vulnerabilities in widely used programs.
Limited Access for Security Organizations
Only 40 security organizations are initially granted access to the advanced AI model.
Company Halts Release Due to Danger
Anthropic has delayed the release of its Claude model due to concerns about potential risk.
Unauthorized Access to AI Model
An unreleased version of the AI model was accessed by unauthorized individuals.
Perspectives
- All sources agree that Anthropic’s new AI model, ‘Mythos,’ is considered dangerous.
- The primary concern is the model’s potential to be exploited by hackers and amplify cyberattacks.
- Anthropic is limiting access to the model due to these significant risks.
- The model’s ability to identify vulnerabilities in software is a key factor in the decision.
Source RTBF, RTL Nieuws, ORF News, and El País state the AI is ‘too dangerous’ and requires restricted access.
RTBF, RTL Nieuws, ORF News, El País
Source Politico EU and Le Monde state Anthropic is working with cybersecurity firms to test defenses and comply with regulations.
Politico EU, Le Monde
Sources FAZ, El Mundo, and Público report limited access to 40-50 security organizations.
FAZ, El Mundo, Público
Source Der Standard and ZEIT Online mention a trial version being shared with large tech companies.
Der Standard, ZEIT Online