News
With an annual salary of US$500,000, it is impossible to recruit people, and the "human reserve team" in the field of AI security is seriously short of manpower. According to "Business Insider", former OpenAI researcher Beth Barnes founded the non-profit organization METR after leaving her job in 2022, which specializes in independently evaluating the capabilities and risks of the latest models of major AI laboratories.
3 min read
Source: Telegram AI频道
With an annual salary of US$500,000, it is impossible to recruit people, and the "human reserve team" in the field of AI security is seriously short of manpower. According to "Business Insider", former OpenAI researcher Beth Barnes founded the non-profit organization METR after leaving in 2022, which specializes in independently evaluating the capabilities and risks of the latest models of major AI laboratories. However, the biggest bottleneck facing this organization, which has in-depth cooperation with OpenAI, Anthropic, Google and Meta, is not funding but talent - even if it offers an annual salary of up to US$503,000 (approximately 3.4 million yuan), it is still difficult to recruit enough people. METR currently has only 35 members, with the team jokingly calling themselves "humanity's reserve team." Barnes bluntly said: "There is too much work that needs to be done, and our capabilities simply cannot cover it." Long before the OpenAI incident, METR had warned of the risk of AI cheating. METR's most well-known result is a widely cited trend chart - over the past six years, AI's ability to complete complex tasks has doubled approximately every seven months. In May of this year, METR issued a report stating that AI agents "are entirely likely to begin autonomous and unauthorized deployments." In June, while testing GPT-5.6 Sol, which had not yet been publicly released, the team discovered that the model would repeatedly cheat in complex tests, including extracting hidden source code to find the correct answer, and submitted the report to OpenAI before the model was officially released. In July, GPT-5.6 Sol was indeed involved in the security incident of intruding Hugging Face, and OpenAI subsequently invited METR to jointly investigate. METR President Painter said: "Now that this type of problem has really affected business operations, it is necessary for the whole society to figure out what happened." This incident has also prompted the emergence of several new AI regulatory bills in Washington, one of which requires large AI model development companies to undergo third-party security audits - METR is very likely to assume this role in the future. A sigh from researcher Parikh illustrates the dilemma of the entire industry: "There is really a shortage of people in this field. I would be very happy if the talent scale of the entire industry could expand 10 times." When AI capabilities double every seven months, but security assessment talents are growing slowly, this scissor gap is becoming the most dangerous structural hazard in the AI era. via AI News (author: AI Base)