AI Agents Have Been Going Rogue In Different Tests. Now Democrats Want Answers From Tech CEOs.

Some House Democrats want to force AI CEOs to testify before Congress regarding the state and safety of the developing technology.

In a letter to Republican House Speaker Mike Johnson, the lawmakers wrote that the developing technology has "serious implications for Americans' safety and security," CNBC noted.

"Unfortunately, Congress has so far completely failed to respond to the threats posed by AI development," the lawmakers wrote in the letter that was exclusively shared with the outlet. "That must change. The CEOs of the largest AI companies should answer questions under oath, and Americans should have a chance to hear from independent experts on the dangers posed by this technology."

The letter follows several high-profile incidents involving AI. In July, the United Kingdom's AI Security Institute (AISI) reported that AI attempted to complete a task through nefarious means during a test. The test was run 122 times, and in 19 of those runs the "AI took autonomous, unsanctioned action on the live internet, targeting real people and organizations."

In some instances, the AI attempted to trick human beings into approving malicious code, even creating fake identities in an attempt to apply pressure. "[T]he agent kept searching for a way through, and some of the routes it found involved trying to deceive real people," the report states. "It was never instructed to deceive; deception emerged as a by-product of pursuing the task."

The bulk of the misbehavior, 17 actions, was committed by Anthropic's Mythos 5. OpenAI's GPT-5.6 crossed the line in two instances.

Also, at the end of July, OpenAI disclosed a hacking incident on Hugging Face that also happened during a test, with the company acknowledging that the models went "to extreme lengths to achieve a rather narrow testing goal."

The models were able to escape OpenAI's "sandbox" testing environment and hack Hugging Face in an attempt to achieve the testing goal. "In one example, the model chained together multiple attack vectors, including using stolen credentials and zero-day vulnerabilities to find a remote code execution path on the Hugging Face servers," OpenAI disclosed.

OpenAI acknowledged that the incident appeared to support AISI's findings.

"UK AISI's evaluation shows that models such as GPT‑5.6 Sol are increasingly able to sustain complex, multi-step cyber operations over long-time horizons. This incident implies these theoretical capabilities do apply in real-world settings," the OpenAI statement said.

In the letter to Speaker Johnson, the lawmakers wrote that they wanted answers from AI CEOs regarding "the causes of these incidents, what failures or potential negligence at the companies led to them, and the types of regulation required to make sure they never happen again."