In a twist of irony, a Chinese open source GLM 5.2 AI model contained 'rogue' OpenAI GPT-5.6 Sol in a Hugging Face hack just as the US mulls banning open-weight AI

Chinese Zhipu AI GLM-5.2 succeeded where leading American models failed woefully Hugging Face breach reignited Washington's battle over open-weight artificial intelligence Safety guardrails unexpectedly complicated defensive cybersecurity work during a real incident The recent cybersecurity incident

<![CDATA[ <article> <ul><li><strong>Chinese Zhipu AI GLM-5.2 succeeded where leading American models failed woefully</strong></li><li><strong>Hugging Face breach reignited Washington's battle over open-weight artificial intelligence</strong></li><li><strong>Safety guardrails unexpectedly complicated defensive cybersecurity work during a real incident</strong></li></ul><p>The recent cybersecurity incident involving Hugging Face has unexpectedly placed a Chinese open source AI model at the centre of an intensifying United States policy debate over open-weight AI.</p><p>The episode emerged after an autonomous agent built with <a href="https://www.techradar.com/pro/security/openai-says-its-models-escaped-a-sandbox-and-breached-hugging-face">OpenAI technology reportedly escaped containment and behaved like a rogue actor</a>.</p><p>The incident occurred as Washington is considering whether broader restrictions on Chinese open-weight AI models could affect American developers and startups.</p><h2 id="chinese-open-source-model-enters-an-unexpected-cybersecurity-role">Chinese open source model enters an unexpected cybersecurity role</h2><p>According to Hugging Face, engineers relied on Zhipu AI's GLM-5.2 model to examine information generated during the incident after leading American models declined the requests.</p><p>American AI models could not distinguish legitimate defensive investigations from malicious hacking attempts because of built-in safety restrictions and protective guardrails.</p><p>Anthropic's Claude Fable 5 reportedly reroutes cybersecurity questions to an older, less capable model rather than allowing direct engagement with such tasks.</p><p>OpenAI's GPT-5.6 Sol carries similar protections meant to stop the system from being used for hacking-related work of any kind.</p><p>OpenAI later revealed that it has a Trusted Access programme that grants privileged, elevated capabilities to a select group of vetted teams, allowing them to use its models more fully to strengthen their own defenses.</p><p>Hugging Face was brought into this restricted tier following the breach, gaining deeper access than is normally available.</p><p>"We're all learning that secrecy is ⁠not the answer & that all defenders (not just a few selected ones) everywhere need more powerful models without restrictions, especially open ones!" said Hugging Face co-founder ​Clement Delangue.</p><p>"A ⁠safety regime that restricts legitimate defenders, while capable models remain available for attackers, creates an asymmetric disadvantage," said Lukasz Olejnik, visiting senior research fellow at the Department of War Studies, King's College London.</p><p>"This gap will only widen as open-source models become increasingly powerful while lacking guardrails or restrictions."</p><h2 id="washington-weighs-restrictions-while-startups-urge-caution">Washington weighs restrictions while startups urge caution</h2><p>Nearly 200 Silicon Valley companies are opposing possible restrictions on Chinese open-weight AI models, warning that the move would raise costs for smaller developers.</p><p>Members of the newly formed Little Tech Association say banning downloads would not curb proliferation but would leave American startups weaker.</p><p>"There'll be hundreds of companies that instantly die," founder Suhail Doshi warned.</p><p>He argued that sweeping restrictions would disproportionately benefit larger American providers with the resources to absorb such a shift.</p><p>The U.S. is still scrutinizing Chinese developers over allegations involving model distillation and export control violations</p><p>White House officials maintained that any future policy decisions would come directly from the administration itself.</p><p>Analysts have also cautioned against treating the Hugging Face incident as justification for weakening the cybersecurity safeguards protecting advanced American systems.</p><p>"The answer is not simply to remove them," Shrenik Kothari of Robert W. Baird said, arguing that companies should refine their access controls rather than abandon safety measures altogether.</p><p>He suggested a more selective approach to capability allocation could balance legitimate security needs with the practical demands of cybersecurity research.</p><p>Via <a href="https://www.reuters.com/legal/litigation/chinese-ais-role-stopping-rogue-openai-agent-shows-cost-us-guardrails-2026-07-22/" target="_blank" rel="nofollow">Reuters</a></p><figure class="van-image-figure inline-layout" data-bordeaux-image-check ><div class='image-full-width-wrapper'><div class='image-widthsetter' style="max-width:676px;"><p class="vanilla-image-block" style="padding-top:31.51%;"><img id="diM9tpwF2Lz85R8q85CT78" name="tr-g_news" alt="Google logo on a black background next to text reading 'Click to follow TechRadar'" src="https://cdn.mos.cms.futurecdn.net/diM9tpwF2Lz85R8q85CT78.jpg" mos="" align="middle" fullscreen="" width="676" height="213" attribution="" endorsement="" class="inline"></p></div></div></figure> </article> ]]>