OpenAI backs Kids Online Safety Act

OpenAI endorsed the Kids Online Safety Act on May 13, adding one of the most prominent artificial intelligence companies to the backers of a bipartisan bill aimed at forcing online platforms to do more to protect minors. Senators Richard Blumenthal, a Connecticut Democrat, and Marsha Blackburn, a Tennessee Republican, announced the endorsement ahead of a Senate hearing on child safety and social media verdicts. (blumenthal.senate.gov) Chris Lehane, OpenAI’s vice president of global affairs, said in the senators’ release that the company was “really excited” to support the bill and called it “a big and important step” for protecting children online. (blumenthal.senate.gov) OpenAI separately has spent recent months publishing child-safety proposals and teen-protection materials that frame its policy push around AI-specific safeguards as well as broader platform rules. The endorsement lands as OpenAI is also expanding the reach of its latest models, including systems the company and outside evaluators say show strong cybersecurity capabilities. That combination — a public call for tougher child-safety rules while advancing more capable models — places the company in two policy fights at once: platform accountability and frontier-model risk. ### Why did OpenAI back this bill now? The May 13 endorsement was released by Blumenthal and Blackburn hours before Blackburn chaired a Senate Subcommittee on Privacy, Technology, and the Law hearing on why recent verdicts should push Congress toward federal action. (blumenthal.senate.gov) Senate Commerce Committee Chairman Ted Cruz also said at a Capitol rally that he would try to move the legislation out of committee and get it signed into law by the end of the year. OpenAI has been building a public child-safety agenda since at least April 8, when it published a Child Safety Blueprint that called for stronger protections against AI-enabled child sexual exploitation and for collaboration with groups including NCMEC, Thorn and state attorneys general. (blumenthal.senate.gov) That document said the company wanted age-appropriate design, clearer standards and broader coordination across industry and government. (deploymentsafety.openai.com) ### What exactly did OpenAI say? Chris Lehane said, “When it comes to tech, we should treat kids as kids and that means making sure it is safe, age-appropriate, and grounded in real-world support.” He said KOSA was “a big and important step toward doing right by our kids.” The company’s recent policy papers use similar language. OpenAI’s April child-safety blueprint described child sexual exploitation as “one of the most urgent challenges of the digital age,” while a separate teen-safety release said the company had called for industry-wide protections and was publishing safety policies for developers. (blumenthal.senate.gov) ### How does this fit with OpenAI’s newest models? OpenAI published its GPT-5.5 system card on April 23 and said it had run the model through predeployment safety evaluations, including targeted red-teaming for advanced cybersecurity and biology capabilities. (openai.com) The company said it gathered feedback from nearly 200 early-access partners before release and updated the card on April 24 with additional information about safeguards for API deployment. On May 7, OpenAI also unveiled GPT-5.5-Cyber in limited preview for vetted defenders responsible for critical infrastructure, according to Politico and OpenAI materials cited there. OpenAI spokespersons told Politico the company had previewed the model for the White House, the Commerce Department’s Center for AI Standards and Innovation and select congressional committees. (blumenthal.senate.gov) Reports this week also pointed to internal references to GPT-5.6 in developer or Codex logs, but those sightings have not been confirmed in any official OpenAI announcement reviewed for this article. (openai.com) The available primary-source record supports that GPT-5.5 and GPT-5.5-Cyber are public; it does not independently confirm a broader GPT-5.6 launch. ### What do outside evaluators say about the cyber risk? The UK AI Security Institute said last week that GPT-5.5 is “one of the strongest models” it has tested on cyber tasks and “the second model to solve one of our multi-step cyber-attack simulations end-to-end.” Bruce Schneier, citing that evaluation, wrote on May 13 that GPT-5.5 appeared comparable to Anthropic’s Claude Mythos in finding software vulnerabilities. (deploymentsafety.openai.com) Politico reported on May 7 that Anthropic had limited Mythos testing to a small group because of its hacking capabilities, while OpenAI said GPT-5.5-Cyber would be available only to vetted members of its Trusted Access for Cyber program. (politico.com) OpenAI said approved users would have to install advanced account security by June 1. ### What happens next in Washington and at OpenAI? The next immediate milestone is in Congress: Blackburn’s May 13 hearing and Cruz’s stated plan to advance KOSA through committee this year. The bill’s prospects will depend on whether Senate leaders and House lawmakers move the measure after the latest round of hearings and negotiations. (deploymentsafety.openai.com) At OpenAI, the next visible checkpoints are already public. The GPT-5.5 system card remains posted on the company’s safety hub, the Child Safety Blueprint is published on OpenAI’s site, and the Trusted Access for Cyber program includes a June 1 account-security requirement for approved users of GPT-5.5-Cyber. (aisi.gov.uk) (deploymentsafety.openai.com) (blumenthal.senate.gov) (politico.com)