{"id":23885,"date":"2025-04-16T05:01:07","date_gmt":"2025-04-16T05:01:07","guid":{"rendered":"https:\/\/www.europesays.com\/uk\/23885\/"},"modified":"2025-04-16T05:01:07","modified_gmt":"2025-04-16T05:01:07","slug":"openai-may-adjust-its-safeguards-if-rivals-release-high-risk-ai","status":"publish","type":"post","link":"https:\/\/www.europesays.com\/uk\/23885\/","title":{"rendered":"OpenAI may ‘adjust’ its safeguards if rivals release ‘high-risk’ AI"},"content":{"rendered":"

OpenAI has updated<\/a> its Preparedness Framework \u2014 the internal system it uses to assess the safety of AI models and determine necessary safeguards during development and deployment. In the update, OpenAI stated that it may \u201cadjust\u201d its safety requirements if a competing AI lab releases a \u201chigh-risk\u201d system without similar protections in place.<\/p>\n

The change reflects the increasing competitive pressures on commercial AI developers to deploy models quickly. OpenAI has been accused of lowering safety standards<\/a> in favor of faster releases, and of failing to deliver timely reports detailing its safety testing<\/a>.\u00a0Last week, 12 former OpenAI employees filed a brief<\/a> in Elon Musk\u2019s case against OpenAI, arguing the company would be encouraged to cut even more corners on safety should it complete its planned corporate restructuring.<\/p>\n

Perhaps anticipating criticism, OpenAI claims that it wouldn\u2019t make these policy adjustments lightly, and that it would keep its safeguards at \u201ca level more protective.\u201d<\/p>\n

\u201cIf another frontier AI developer releases a high-risk system without comparable safeguards, we may adjust our requirements,\u201d wrote OpenAI in a blog post<\/a> published Tuesday afternoon. \u201cHowever, we would first rigorously confirm that the risk landscape has actually changed, publicly acknowledge that we are making an adjustment, assess that the adjustment does not meaningfully increase the overall risk of severe harm, and still keep safeguards at a level more protective.\u201d<\/p>\n

The refreshed Preparedness Framework also makes clear that OpenAI is relying more heavily on automated evaluations to speed up product development. The company says that while it hasn\u2019t abandoned human-led testing altogether, it has built \u201ca growing suite of automated evaluations\u201d that can supposedly \u201ckeep up with [a] faster [release] cadence.\u201d<\/p>\n

Some reports contradict this. According to the Financial Times<\/a>, OpenAI gave testers less than a week for safety checks for an upcoming major model \u2014 a compressed timeline compared to previous releases. The publication\u2019s sources also alleged that many of OpenAI\u2019s safety tests are now conducted on earlier versions of models rather than the versions released to the public.<\/p>\n

In statements, OpenAI has disputed the notion that it\u2019s compromising on safety.<\/p>\n

\n

OpenAI is quietly reducing its safety commitments.<\/p>\n

Omitted from OpenAI\u2019s list of Preparedness Framework changes:<\/p>\n

No longer requiring safety tests of finetuned models https:\/\/t.co\/oTmEiAtSjS<\/a><\/p>\n

\u2014 Steven Adler (@sjgadler) April 15, 2025<\/a><\/p>\n<\/blockquote>\n

Other changes to OpenAI\u2019s framework pertain to how the company categorizes models according to risk, including models that can conceal their capabilities, evade safeguards, prevent their shutdown, and even self-replicate. OpenAI says that it\u2019ll now focus on whether models meet one of two thresholds: \u201chigh\u201d capability or \u201ccritical\u201d capability. <\/p>\n

OpenAI\u2019s definition of the former is a model that could \u201camplify existing pathways to severe harm.\u201d The latter are models that \u201cintroduce unprecedented new pathways to severe harm,\u201d per the company.<\/p>\n

\u201cCovered systems that reach high capability must have safeguards that sufficiently minimize the associated risk of severe harm before they are deployed,\u201d wrote OpenAI in its blog post. \u201cSystems that reach critical capability also require safeguards that sufficiently minimize associated risks during development.\u201d<\/p>\n

The updates are the first OpenAI has made to the Preparedness Framework since 2023.<\/p>\n