Archive — US — techcrunch
OpenAI may ‘adjust’ its safeguards if rivals release ‘high-risk’ AI
OpenAI is revising its Preparedness Framework to potentially adjust safety requirements if a competitor releases a 'high-risk' AI without adequate safeguards. This decision reflects competitive pressures to accelerate AI development. Despite accusations of lowering standards for quicker releases, OpenAI asserts that any adjustments will prioritize safety, using more automated evaluations while still requiring human oversight.
Key points
- OpenAI updates its Preparedness Framework to enhance AI safety amidst competitive pressures.
- The company may lower safeguards if competitors release high-risk AI systems without adequate protections, while ensuring overall risks do not increase.
- Automated evaluations will speed up development, but concerns arise over the reduced testing timelines and alleged compromises on safety.
- Recent allegations suggest OpenAI has shifted away from robust human-led testing, sparking criticism from former employees and regulators.
- The new framework distinguishes between 'high' and 'critical' capability models, each with tailored safeguards.
Why it matters
- The shift in OpenAI's safety policies aligns with the broader trend of rapid AI deployment seen in the tech industry, raising concerns about regulatory oversight.
- In previous years, tech companies faced backlash for prioritizing speed over safety, leading to calls for more stringent regulations.
- The growing use of automated evaluations reflects a shift in tech development strategies, matching 2025's focus on efficiency and scalability in AI deployments.
OpenAI's framework adjustments highlight the tension between rapid AI development and essential safety measures. As industry pressure mounts, it remains crucial for companies to balance innovation with responsibility to prevent severe risks.