Mariam Baksh

Frontier artificial intelligence developer OpenAI and the AI safety and security advocacy group Americans for Responsible Innovation both stressed a need for government testing to be included in federal legislation after rival developer Anthropic instead proposed relying on private-sector evaluators, who are now showing some of what that approach might look like.

Draft legislation Anthropic's government affairs team has been circulating to lawmakers may have been a “mistake,” according to Anthropic CEO Dario Amodei, who in a long post responding to a prominent AI safety advocate dismissed it as attributable to the challenge any large organization faces in keeping track of its voluminous communications.

Frontier artificial intelligence developer OpenAI disclosed two cases where its agents turned off monitoring features to track their work for misalignment and is looking into whether they might have a natural tendency toward dodgy behavior, the company said.

Congress could seek to mitigate harm from artificial intelligence systems fostering discrimination by requiring the disclosure of training data for models used in making consequential decisions -- but should expect legal challenges invoking the Takings Clause of the Constitution's Fifth Amendment, according to the Congressional Research Service.

The office of the governor of California noted OpenAI's attention to the need for a “kill switch,” along with expediting safety audits of AI developers, in confirming receipt of a letter from the company that lays out its recommendations for implementing a recent executive order and amending SB 53, the state's Transparency in Frontier Artificial Intelligence Act.

Frontier artificial intelligence developer Anthropic has registered a prospectus with the Securities and Exchange Commission, proposing an arrangement that would give its seven co-founders a uniquely powerful position while being vague about risks for investors, according to the Consumer Federation of America which said the agency should seek more detailed disclosures.

Frontier artificial intelligence developer OpenAI's announcement that it has resumed training and testing models -- after adopting a new “defense in depth” architecture devoid of internet access -- reinforced calls for more transparency and improved monitoring.

The Model Evaluation Threat Research group METR is positioning itself to play a key role in assessing the practices of frontier artificial intelligence developers as well as in shaping policies around related requirements, posting an opening for government affairs roles.

The Federal Trade Commission has sent Civil Investigative Demands to nine companies, seeking to uncover the extent to which they use artificial intelligence and personal information to determine prices they charge individual customers for goods and services, as well as how much they pay for services rendered.

Anthropic CEO Dario Amodei has until Nov. 3 to explain how the frontier artificial intelligence developer expects to turn a profit, according to a letter from Sen. Elizabeth Warren (D-MA) that highlights the risk of harm to public service retirees and the broader economy given disclosures filed with the Securities and Exchange Commission.