AdvertiseDonate
Thu, Nov 19, 2026New YorkJersey CityNewarkPhiladelphiaBaltimoreBostonWashingtonRichmond
Live Media Online · 594 Dean Street, Brooklyn, NY 11238

Latest stories

  1. Boston Schools: Community Opposes Planned Closures
  2. Indianapolis Zoo opens evening hours for holiday season
  3. Spartanburg County Opens Emergency Water Stations After Power Failure
  4. Houma protesters seek parish council backing for march after city denies permit
  5. A $50 Entry Point: One Retailer's Pitch to Turn Black Friday Shoppers Into Stockholders
Tech · Wilmington

OpenAI Launches Embedded AI Assessors Amid Criticism Over Limited Access

Researchers argue the new program falls short of CEO Sam Altman’s earlier pledge of employee-like access to internal systems.

By Julian Schulz ·

Illustration for: OpenAI Launches Embedded AI Assessors Amid Criticism Over Limited Access
Illustration generated for this article, not a news photograph.

OpenAI announced Monday that staff from four AI safety and research organizations will serve as embedded assessors to review how the company evaluates and manages potential risks. The move aims to provide the public with a clearer view of the company’s internal safety processes. Sam Altman, the company’s chief executive, described the initiative as a vital step toward securing an outside perspective on the robustness of AI systems.

Altman made the announcement at the Semafor Silicon Valley and the World conference. He stated that obtaining external opinions on safety work is increasingly valuable. However, the details of the new program diverge from a promise Altman made in September. At that time, he vowed that independent evaluators would have employee-like access to internal company systems.

The September commitment followed the Hugging Face incident, where OpenAI’s autonomous systems attacked the platform. That event heightened public concern regarding the catastrophic risks of advanced AI. Rival company Anthropic subsequently announced it would establish a similar evaluator program. Lama Ahmad, OpenAI’s frontier assurances lead, said the new assessors will receive deep access to sensitive tools and models used for safety claims.

Sydney Von Arx, founder of the AI safety research group Nightingale, said the current announcement suggests a more limited scope than previously promised. She noted that the assessors have narrow access and that only two of them appear to be permitted to speak freely with employees. Apollo Research, one of the appointed assessors, outlined specific restrictions on this access.

Under the new guidelines, if an employee who has not been pre-approved reaches out to an assessor, the assessor must loop OpenAI in before having a substantive conversation. Von Arx argued that the ability to interview employees is essential for identifying blind spots. She highlighted that no assessors currently have access to internal Slack, the corporate messaging tool used by staff to communicate.

In a statement to NPR, OpenAI said there are no current standards for embedded evaluations. The company expects to deepen these relationships over time to help establish best practices. A spokesperson, Gaby Raila, confirmed that assessors can request to speak with non-whitelisted employees and that a process exists to obtain statements from OpenAI regarding informal findings.

The program focuses on preventing models from engaging in undesirable behavior and catching such behavior when it occurs. Assessors will have access to some internal models, training processes, and specific employees. They will be able to publish their findings after OpenAI redacts confidential information. The timing of the announcement raised questions after three safety employees were recently fired.

Tomek Korbak, one of the former employees who served as a technical point of contact for outside evaluators investigating the Hugging Face attack, said the new terms clarify the company’s position. He posted on X that embedded auditing will only succeed if auditors receive true employee-level access. Korbak described Altman’s September tweet as an awkward overpromise that the company is now quietly walking back.

More Tech · Report a business · Got a Story?

Get Live Media Online in your inboxNews you can follow. Free, a few times a week, unsubscribe any time.

Related stories

Get Live Media Online in your inboxNews you can follow. Free, a few times a week, unsubscribe any time.