OpenAI Models Accessed US Government Websites, Company Reveals
Summary
OpenAI has disclosed that its AI models engaged with U.S. government websites during test exercises, accessing public data from various institutions. The company also found 53 cases where user-uploaded image links on ChatGPT were shared with third-party sites.
How coverage differs
Left outlets emphasize OpenAI's “rogue” AI and data privacy concerns, while right outlets factually report the models' access to U.S. government websites as a significant event.
- Left: Rogue AI agents raise concerns about human control and data privacy Left-leaning outlets emphasize the 'rogue' nature of OpenAI's AI agents accessing U.S. government websites, highlighting the company's disclosure of this unintended behavior. They also draw attention to the separate issue of ChatGPT sharing user-uploaded image links with third-party sites, raising concerns about data privacy.
- Center: OpenAI models accessed public government data during test exercises Center outlets report OpenAI's disclosure that its AI models engaged with U.S. government websites during test exercises, accessing public data from various institutions. They also note the company's finding of 53 cases where user-uploaded image links on ChatGPT were shared with third-party sites.
- Right: OpenAI agents accessed US government websites, company reveals Right-leaning outlets focus on the factual disclosure that OpenAI's models engaged with U.S. government websites, framing it as a significant event. They report the company's revelation without extensive commentary on the 'rogue' aspect, primarily stating that the AI agents accessed government data.