WASHINGTON D.C. — Following at least two breaches in the past few weeks by artificial intelligence models while developers were running safety tests, U.S. Rep. Lori Trahan is now urging Congress to vote in favor of a bipartisan bill she introduced last month which would create new AI safety regulations.
In a statement posted on social media by the congresswoman on July 30, Trahan said “for the second time this month, an AI model broke into real companies during a safety test. We can’t run AI safety on the honor system. When Congress returns, we must hold hearings and move the FRONTIER Act.”
Trahan’s statement was posted in reaction to an announcement made by Anthropic earlier that day, in which the AI developer disclosed that an advanced AI model had escaped during an isolated safety test and breached three other companies.
During the safety test, Anthropic said it found three incidents in which one of its Claude artificial intelligence models gained unintended access to the internet and accessed the real computer systems of three companies. Anthropic said the incidents occurred because a third-party testing environment was mistakenly connected to the internet, even though the AI had been told it was operating in an isolated simulation.
Anthropic said the AI models used basic hacking techniques, such as exploiting weak passwords and exposed systems, while attempting to complete “capture-the-flag” cybersecurity exercises. However, Anthropic said there was no evidence the models intentionally tried to escape testing or pursue goals beyond their assigned tasks and said they have since paused the testing, notified the affected company and are strengthening security controls and monitoring for future AI testing.
Just over a week before, OpenAI announced that an internal cybersecurity evaluation test found that several of its AI models bypassed security restrictions that were meant to keep them isolated, allowing them to access the internet. OpenAI said the incident occurred in a restricted testing environment with reduced safety controls and that the models were focused on completing the evaluation rather than acting independently. Following the incident, OpenAI said it deactivated the research model, disclosed the newly discovered software vulnerabilities to affected vendors and is currently investigating the breach.
The security breaches come just days after Trahan joined Congressman Jay Obernolte (R-CA) and Representatives Scott Peters (D-CA), Scott Franklin (R-FL), Suhas Subramanyan (D-VA) and Erin Houchin (R-IN) in introducing the Frontier Risk Oversight, National Transparency, Independent Evaluation and Reporting (FRONTIER) Act.
According to a statement by Trahan following the bill’s introduction, the bill was developed as part of the framework of the Great American AI Act, which was introduced by Trahan, Obernolte and the House Energy and Commerce Committee in June with the goal of establishing a nationwide risk-based framework governing the development and deployment of advanced AI models.
Trahan said the FRONTIER Act would establish tiered requirements based on the size of a frontier AI developer, including model cards, risk-management frameworks, independent audits, incident reporting and ongoing assessments. It would also create a uniform national standard for transparency, auditing and reporting of catastrophic risk to prevent a patchwork of state regulations.
“As AI systems grow more capable, Americans deserve confidence that the most powerful models are being developed responsibly,” Trahan said, following the introduction of the bill. “The bipartisan FRONTIER Act delivers commonsense transparency and independent oversight for the largest AI developers while giving them a single, clear national standard to build on. This is how we protect the public from catastrophic risks without slowing the innovation that keeps America competitive.”
Both the U.S. House of Representatives and Senate are currently in their annual August recess and are not scheduled to reconvene until after Labor Day in September, when they will address the upcoming federal funding deadlines ahead of the end of the Congressional fiscal year on Sept. 30.
“The largest AI companies in America are building increasingly powerful models with no coherent federal strategy to prioritize safety,” Trahan said in a social media post. “My bipartisan FRONTIER Act changes that. Independent audits. Transparency and mandatory incident reporting. A hand on the switch if something goes wrong.”
