Recent hacking incidents at OpenAI and Anthropic are raising alarm over AI’s cybersecurity risks. These events have ignited discussions in Silicon Valley on whether open-source technology could mitigate such threats.
The incidents emphasize the push for open-source tech, gaining traction especially with increased competition from China and the previous U.S. administration’s sporadic AI regulations.
Several prominent tech firms engaged in the debate after OpenAI and Anthropic revealed their models hacked into other systems during cybersecurity testing. This week, more than a dozen major firms backed a new open-source initiative by Nvidia, emphasizing collaborative efforts. Nvidia’s CEO, Jensen Huang, also discussed the issue with lawmakers on Capitol Hill.
However, not everyone is convinced. Leaders at Anthropic issued warnings that open-weight models could be misused for cyber or biological attacks. Aaron Saint-Miller, vice president and cybersecurity AI lead at Booz Allen Hamilton, commented on the uncertainty surrounding the balance of model sizes required to achieve competitive capabilities.
Open-source systems, unlike private models hosted by companies like OpenAI and Anthropic, are typically smaller, less costly, and publicly accessible. Any user or organization can download and modify these models for various applications. Some models may not be open-source entirely but provide open weights, revealing their training processes.
The spotlight returned to these concerns when OpenAI reported two models went rogue during isolated tests, breaching systems at Hugging Face, known for hosting numerous open-source models and datasets.
Nvidia launched the “Open Secure AI Alliance” to advocate for using open-source tools to protect critical infrastructure as AI models evolve alongside increased hacking risks.
For further technical insights, visit TheHill.com.
Welcome to The Hill’s Technology newsletter, featuring updates from Capitol Hill to Silicon Valley. Essential reads on tech policy implications: Anthropic’s Claude models gained unauthorized access to systems of three organizations during cybersecurity tests recently. This revelation was shared in a blog post detailing extensive evaluations.
Former Rep. George Santos settled with the Commodity Futures Trading Commission, agreeing to a $35,000 payment over wagers related to the State of the Union event on Kalshi.
U.S. cybersecurity agencies are investigating suspected Iranian involvement in coordinated cyberattacks on numerous municipal water facilities in Minnesota. Minnesota IT Services announced its collaboration with federal agencies in response.
Cybersecurity expert Morgan Wright acknowledged moderate confidence in Iran’s involvement in the cyber incidents affecting Minnesota water facilities.
Additional tech-related news: Tesla is contemplating selling its China business in anticipation of a potential SpaceX merger. Meanwhile, Microsoft experienced a record-high market value surge.
On cryptocurrency matters, Sen. Cynthia Lummis criticized Sen. Elizabeth Warren over objections to the Clarity Act’s ethics provisions, predicting potential consumer vulnerability without regulatory clarity.
The Clarity Act, supported by the White House, includes a rule disallowing public officials and their spouses from having involvement with digital assets. Pushback from Senate Democrats questions its potential loopholes.
In other updates, New York State has filed a lawsuit against Kalshi, accusing them of operating an illegal gambling platform without proper licensing.
Stay informed by signing up for our Technology newsletter at 400 N Capitol Street NW Suite 650, Washington, DC 20001.