Anthropic’s New AI Model Sparks Safety Concerns

Anthropic’s latest AI model, Mythos, is causing safety and cybersecurity concerns. Matt Shumer, co-founder and CEO of OthersideAI, outlined these issues on ‘The Sunday Briefing.’ Mythos has shown the capability to identify software vulnerabilities autonomously. This ability led Anthropic to limit public access to the model.

In a precautionary move, Anthropic has given early access to major companies and the US government. This step aims to bolster cyber defense against potential threats to critical infrastructure and national security.

Shumer explained, “Anthropic is taking proactive measures by working with key stakeholders to ensure cybersecurity threats are mitigated.”

The Tarbell Center for AI Journalism has been placing reporter fellows with reputable media outlets. These outlets include Bloomberg, Time, NBC News, The Guardian, and the Los Angeles Times. However, disclosure of these relationships varies among the outlets.

Tarbell ensures its fellows receive fair compensation. Salaries range between $60,000 to $80,000, with experienced fellows earning up to $110,000. The organization supports independent journalism and promotes accountability in AI technology discussions.

A controversy involves the lack of consistent disclosure. Not all articles reveal the fellow’s support from Tarbell, leading to transparency questions. For example, Time did not disclose support for reporter Harry Booth in his coverage of Anthropic’s model.

The Verge also did not mention Robert Hart’s fellowship in his article about OpenAI’s security disputes with Anthropic.

Some publications, like NBC News, have opted to add editorial notes. These notes clarify the Tarbell fellowship’s role in specific articles. They ensure readers understand the funding context and maintain transparency.

Tarbell receives funding from several organizations, including Coefficient Giving and the Future of Life Institute. Coefficient Giving has supported Tarbell with over $4.7 million in grants. The funding aims to advocate for AI safety standards.

Tarbell emphasizes transparency in its operations. They require fellows and media outlets to disclose Tarbell’s support prominently. As of now, there are six active fellows reporting, all of whom make these disclosures.