OpenAI has created a brand new Security and Safety Committee lower than two weeks after the corporate dissolved the workforce tasked with defending humanity from AI’s existential threats. This newest iteration of the group accountable for OpenAI’s security guardrails will embody two board members and CEO Sam Altman, elevating questions on whether or not the transfer is little greater than self-policing theatre amid a breakneck race for revenue and dominance alongside accomplice Microsoft.
The Security and Safety Committee, shaped by OpenAI’s board, might be led by board members Bret Taylor (Chair), Nicole Seligman, Adam D’Angelo and Sam Altman (CEO). The brand new workforce follows co-founder Ilya Sutskever’s and Jan Leike’s high-profile resignations, which raised quite a lot of eyebrows. Their former “Superalignment Team” was solely created final July.
Following his resignation, Leike wrote in an X (Twitter) thread on Might 17 that, though he believed within the firm’s core mission, he left as a result of the 2 sides (product and security) “reached a breaking point.” Leike added that he was “concerned we aren’t on a trajectory” to adequately tackle safety-related points as AI grows extra clever. He posted that the Superalignment workforce had not too long ago been “sailing against the wind” throughout the firm and that “safety culture and processes have taken a backseat to shiny products.”
A cynical take can be that an organization targeted totally on “shiny products” — whereas making an attempt to fend off the PR blow of high-profile security departures — would possibly create a brand new security workforce led by the identical folks dashing towards these shiny merchandise.
The protection departures earlier this month weren’t the one regarding information from the corporate not too long ago. It additionally launched (and shortly pulled) a brand new voice mannequin that sounded remarkably like two-time Oscar Nominee Scarlett Johansson. The Jojo Rabbit actor then revealed that OpenAI Sam Altman had pursued her consent to make use of her voice to coach an AI mannequin however that she had refused.
In a press release to Engadget, Johansson’s workforce mentioned she was shocked that OpenAI would solid a voice expertise that “sounded so eerily similar” to her after pursuing her authorization. The assertion added that Johansson’s “closest friends and news outlets could not tell the difference.”
OpenAI additionally backtracked on nondisparagement agreements it had required from departing executives, altering its tune to say it wouldn’t implement them. Earlier than that, the corporate compelled exiting workers to decide on between with the ability to communicate in opposition to the corporate and holding the vested fairness they earned.
The Security and Safety Committee plans to “evaluate and further develop” the corporate’s processes and safeguards over the following 90 days. After that, the group will share its suggestions with the whole board. After the entire management workforce critiques its conclusions, it’ll “publicly share an update on adopted recommendations in a manner that is consistent with safety and security.”
In its weblog submit saying the brand new Security and Safety Committee, OpenAI confirmed that the corporate is at the moment coaching its subsequent mannequin, which is able to succeed GPT-4. “While we are proud to build and release models that are industry-leading on both capabilities and safety, we welcome a robust debate at this important moment,” the corporate wrote.