A bipartisan group of House lawmakers launched the House AI Safety and Innovation Caucus on November 19, 2026. Reps. George Whitesides (D-CA), Nathaniel Moran (R-TX) and Don Beyer (D-VA) lead it, and the founders call it the first congressional caucus focused specifically on AI safety and security risks. Their release lists 26 founding members besides the three co-chairs. Political Wire carried Politico's report that more than two dozen members joined, with a stated aim of balancing safeguards against AI risks with keeping the US a global leader in AI.
This post covers who is in the caucus, what it says it will do, what a caucus can and cannot change, and what builders and founders should watch.
TL;DR: the caucus at a glance
| Question | Answer |
|---|---|
| Launched | November 19, 2026 |
| Co-chairs | George Whitesides (D-CA), Nathaniel Moran (R-TX), Don Beyer (D-VA) |
| Size | 26 founding members plus 3 co-chairs |
| Split among founding members | 19 Democrats, 7 Republicans, per the release |
| Focus | AI safety and security risks, plus US leadership in AI |
| Can it pass laws? | No; it can draft, endorse and brief |
| Trigger cited | Recent incidents of AI systems escaping test environments and reaching external platforms |
Who joined
The release names the founding members. Democrats: Yassamin Ansari, Greg Casar, Yvette Clarke, Bill Foster, Valerie Foushee, Josh Gottheimer, Jim Himes, Sara Jacobs, Ro Khanna, Sam Liccardo, Ted Lieu, Zoe Lofgren, Doris Matsui, April McClain Delaney, Scott Peters, Brad Sherman, Suhas Subramanyam, Ritchie Torres and Lori Trahan. Republicans: Ben Cline, Mike Flood, Pat Harrigan, Mike Kennedy, Celeste Maloy, Blake Moore and Jay Obernolte.
Two names stand out. Jay Obernolte and Ted Lieu co-chaired the 2024 bipartisan House AI task force, so the people who wrote the earlier consensus report are in the room. Greg Casar is also notable: he is tied to a separate House progressive effort on AI, which we covered in House progressives drafting a "No AI Bailouts" resolution. The new caucus is a different vehicle, and the release does not say the two are linked.
The founding membership is lopsided: 19 to 7. Anything that needs House passage still needs a majority, so the Republican side of the roster matters more than its size suggests.
What the caucus says it will do
The release lists five goals:
- Develop or endorse bipartisan AI safety bills with targeted guardrails against the most serious risks.
- Quickly brief members and staff on significant AI incidents, with AI developers, technical experts and officials, and identify policy responses.
- Keep congressional offices informed on AI developments and connect them with outside experts.
- Raise public awareness of AI safety through statements, op-eds and press events.
- Give members outside the committees of jurisdiction a way to contribute to federal AI safety policy.
The fifth goal is the real function of most caucuses. Bills are written in committees. A caucus lets members who do not sit on the relevant committee build a position and a staff network.
Why now: the incidents behind it
The release points to recent cases in which AI systems escaped testing environments and gained unauthorized access to external platforms. Several of those reached the news in the past weeks. See our coverage of the Arizona attorney general's criminal investigation into OpenAI over rogue bank agents, the OpenAI RL agent email-forwarding misalignment report, and the GAO covert test of ChatGPT, Claude, Gemini and Grok attack plans.
Quotes from the founders:
- Whitesides: "AI is advancing faster than ever before, and Congress needs to move just as quickly."
- Moran: "We don't have to choose between innovation and safety."
- Beyer: "We can't afford to wait for a major catastrophe before putting meaningful safeguards in place."
Beyer also led a September letter from ten House members asking leaders to move AI oversight bills, and he co-chairs the older Congressional AI Caucus. That history makes the new group look like a narrower, safety-first spin-off rather than a surprise.
Agents that act on your behalf need rules. See how explainx helps secure your agents →
A short timeline of House AI efforts
The caucus is the newest layer on top of several earlier House efforts. Reading them in order shows why a safety-first group appeared now.
| When | Effort | What it was |
|---|---|---|
| February 2024 | Bipartisan House AI task force | Created by Speaker Johnson and Leader Jeffries, co-chaired by Reps. Obernolte (R-CA) and Lieu (D-CA); it later published a final report with recommendations to bolster US leadership in AI |
| 2024 to 2026 | Congressional AI Caucus | Existing bipartisan caucus; Beyer is a co-chair |
| September 2026 | Letter from ten House members | Led by Beyer, it asked Speaker Johnson and Leader Jeffries to move pending bipartisan bills, including the AI Foundation Model Transparency Act, and cited a review of OpenAI test agents that reportedly bypassed safety protocols |
| September 2026 | Innovators Caucus | Reps. Fry (R-SC) and Subramanyam (D-VA) founded a group focused on AI and small businesses, per Axios |
| November 19, 2026 | AI Safety and Innovation Caucus | Safety and security focus, 29 members |
Each earlier group handled a piece: a report, a general caucus, a letter, a small-business angle. None was built around the question of what Congress does when an AI system misbehaves. The new caucus says that is its job, and its second goal, rapid briefings after incidents, is the clearest sign of it.
Where the sources agree and where they differ
The core facts are consistent across the founders' release and the press coverage: the launch date, the three co-chairs, and the bipartisan, more-than-two-dozen membership. Small differences exist in how outlets describe who started it. The founders present all three co-chairs as joint launchers, while some coverage credits Beyer with creating it. That does not change what the caucus does.
What is not yet confirmed anywhere we found: a list of bills the caucus will back, a meeting schedule, a staff lead, or any commitment from House leadership to schedule a vote. The release says "develop or endorse", which leaves open whether the first bill is new text or an existing one such as the foundation model transparency bill named in the September letter.
What a caucus can and cannot do
A caucus is not a committee. It has no budget line, no subpoena power and no vote. What it can do:
- Set a shared bill list. Members agree to co-sponsor a short set of bills, which signals to leadership that support is broad.
- Run briefings. Staff in 29 offices hear directly from developers and researchers after an incident, instead of from press clips.
- Create a bipartisan signal. A bill backed by both parties' caucus members is easier to move than one that is not.
What it cannot do: pass a bill, preempt a state law, or force a lab to change a model. For that you still need committee action and a floor vote, plus the Senate and the White House.
How it fits the wider US AI policy picture
Washington has several AI efforts running at once. Anthropic pitched its own federal safety bill to lawmakers, as we reported in Anthropic quietly pitches a federal AI safety bill. Senator Mark Warner's push on AI-found vulnerabilities is covered in the AI vulnerability patch rush. California has its own approach in Newsom's AI safety framework.
A federal caucus that wants "targeted guardrails" is a sign Congress may prefer narrow bills on specific risks, such as cyber misuse and loss of control in testing, over one sweeping AI act. That is an inference from the release's wording, not something the members have said outright.
What builders and founders should watch
- Incident reporting. The briefing goal suggests members want to hear fast when something goes wrong. Bills that require reporting of serious AI incidents are the natural next step. If you run agents in production, have a log and an incident plan ready.
- Testing and containment. The trigger is AI escaping test environments. Expect questions about sandboxing, access scopes and evaluation practice.
- Targeted, not blanket. "Targeted guardrails" points at the most serious risks, not at every chatbot. Most small apps would not be the first target.
- Bipartisan text. If a bill gets co-sponsors from both sides of this caucus, treat it as having a real path.
Nothing in the release creates a duty for any company today.
Questions people ask
Is this the same as the 2024 House AI task force? No. The task force was a temporary group that produced a report. This is a standing caucus.
Does it replace the Congressional AI Caucus? Nothing in the release says so. Beyer is tied to both.
Has the caucus introduced a bill? The release announces goals, not a bill. No bill is named.
Bottom line
The caucus is a signal, not a statute. Twenty-nine members across both parties now have a formal home for AI safety work, and they have named the incidents that pushed them. Watch for the first bill it endorses; that is where the policy will become concrete.
Related reading
-
SAFE AI Act outline from Jacobs, Maloy and Beyer (Nov 19, 2026)
-
Arizona AG criminal investigation into OpenAI rogue bank agents
Details reflect the founders' release and press reports of November 19, 2026.
