<p id="speakable-summary" class="wp-block-paragraph">Much has been made of the incident in which one of OpenAI’s agents broke out of its sandboxed test environment and <a href="https://techcrunch.com/2026/07/29/the-hugging-face-ai-break-in-as-told-through-an-increasingly-committed-bear-metaphor/" target="_blank" rel="noreferrer noopener">proceeded to hack</a> the AI hosting platform Hugging Face. OpenAI has since <a href="https://openai.com/index/hugging-face-model-evaluation-security-incident/" target="_blank" rel="noreferrer noopener nofollow">launched an investigation</a> into how the incident occurred, which is still ongoing.</p>
<p class="wp-block-paragraph">Now, anonymous sources have told <a href="https://www.reuters.com/business/openai-finds-evidence-other-ai-agents-escaped-containment-it-widens-hacking-2026-07-31/" target="_blank" rel="noreferrer noopener nofollow">Reuters</a> that more of OpenAI’s agents are believed to have escaped their sandboxes. However, one source downplayed the severity, saying that with those escapes, the agents didn’t appear to leave OpenAI’s network to hack into another company’s. TechCrunch reached out to OpenAI for more information.</p>
<p class="wp-block-paragraph">AI programs acting in bizarre ways has apparently become a weird, almost bragging point for companies. The same week, Anthropic also announced that it <a href="https://therecord.media/anthropic-ai-hacked-three-real-companies" target="_blank" rel="noreferrer noopener nofollow">had discovered not one, but three instances</a> in which its agents had escaped test environments and hacked other organizations.</p>
<p class="wp-block-paragraph">AI companies <a href="https://www.businessinsider.com/anthropic-says-claude-models-went-rogue-hacked-3-companies-testing-2026-7" target="_blank" rel="noreferrer noopener nofollow">have also been accused</a> of using such incidents for marketing purposes — as they generate considerable attention and may underscore how powerful the companies’ products are. The flip side of that is that these disclosures are also <a href="https://www.cnbc.com/2026/07/23/open-ai-hugging-face-hack-kill-switch-bill-congress.html" target="_blank" rel="noreferrer noopener nofollow">ramping up discussions</a> of government regulations.</p>
© 2026 Winzheng.com 赢政天下 | 转载请注明来源并附原文链接