Under Pressure: OpenAI cuts safety testing from months to days amid fierce competition

OpenAI‘s rapid acceleration of safety testing timelines signals a concerning shift in the AI industry’s approach to responsible development. What was once a months-long evaluation process has shrunk to mere days, potentially compromising the thoroughness needed to identify and mitigate harmful AI capabilities before deployment. This transformation highlights the growing tension between competitive market pressures and responsible AI safeguards in an environment where regulatory frameworks remain incomplete.

The big picture: OpenAI has dramatically compressed its safety testing timeline from months to days, according to eight staff members and third-party testers who spoke to the Financial Times.

• The evaluators report having “just days” to complete safety assessments that previously would have been allocated “several months.”
• Sources described the accelerated testing schedule as “reckless” and “a recipe for disaster,” particularly as the company prepares to release its new o3 model, reportedly as soon as next week.

Why this matters: Thorough safety evaluations are critical for identifying potential model risks, such as whether users could jailbreak an AI system to extract harmful capabilities like instructions for creating bioweapons.

• For comparison, sources told FT that OpenAI gave testers six months to review GPT-4 before its release—and concerning capabilities were only discovered after two months of testing.
• The compressed timeline raises serious questions about whether harmful capabilities in new models will be properly identified and mitigated before public release.

Behind the numbers: OpenAI’s shift toward accelerated testing appears driven by competitive pressures, particularly as open-weight models from companies like Chinese AI startup DeepSeek gain market traction.

• Sources indicated that the testing timeline for the upcoming o3 model (the full version of the already-released o3-mini) has been compressed to under a week.
• According to one tester currently evaluating o3, “We had more thorough safety testing when [the technology] was less important.”

The regulatory vacuum: The timeline compression underscores the absence of mandated AI safety standards, as government regulation of AI model development and deployment remains largely nonexistent.

• While companies including OpenAI previously signed voluntary agreements with the Biden administration to conduct routine testing with the US AI Safety Institute, these commitments have reportedly faded under the Trump administration.
• The EU AI Act will eventually require companies to risk test their models and document results, but implementation remains in progress.

What they’re saying: Johannes Heidecke, head of safety systems at OpenAI, defended the company’s approach in comments to the Financial Times.

• “We have a good balance of how fast we move and how thorough we are,” Heidecke told FT.
• However, testers themselves expressed alarm about the process, noting additional concerns including evaluating less-advanced versions of models than those eventually released to the public, or referencing an earlier model’s capabilities rather than fully testing the new one.