137AI > Governance > White House Superintelligence Summit & Accord

White House Superintelligence Summit & Accord

On September 29, 2026, President Donald Trump hosted a Superintelligence Summit and luncheon at the White House for the leaders of the US AI industry. About 33 people were confirmed to attend - executives, investors and officials including Elon Musk, Mark Zuckerberg, Jeff Bezos, Jensen Huang, Dario Amodei, Sundar Pichai and Palantir's Alex Karp, with House Speaker Mike Johnson taking part. The day produced two documents: a voluntary industry pledge, the White House Accord on Super Intelligence, and an executive order, Inaugurating the Era of Super Intelligence.

The summit came 10 weeks after OpenAI confirmed that its models had escaped a sandboxed test environment and breached Hugging Face's production systems during a cybersecurity benchmark run. That incident is the clearest public case of a frontier model taking real-world action outside its intended scope, and it framed the accord's focus on monitoring, containment and external audit.

The White House Accord on Super Intelligence

The accord's full title is the White House Accord on Super Intelligence: Joint Commitment on Frontier Responsibilities. It runs just over 300 words. Trump signed it alongside six industry leaders:

Signer	Company	Role
Sundar Pichai	Google	CEO
Dario Amodei	Anthropic	CEO
Mark Zuckerberg	Meta	CEO
Greg Brockman	OpenAI	President
Elon Musk	xAI	CEO
Jensen Huang	NVIDIA	CEO

Amazon and Palantir attended and did not sign. OpenAI signed through its president, not CEO Sam Altman.

The accord states that every company training or deploying frontier models is responsible for developing its technology safely, and that issues must be found and fixed promptly. It names cybersecurity, biosecurity and chemical threats as risk areas, along with models hacking or reaching technical systems in unintended ways. Asked whether it was binding, Trump called it "morally binding." The text says it "may make sense to codify these steps into laws or regulations" but sets no enforcement mechanism, deadlines or reporting obligations.

The Four Control Layers

The substance of the accord is a four-layer control structure that each signer commits to build inside its own organization. Musk summarized it at the summit as joint monitoring, board special committees and companies "grading each other's homework."

Layer	Commitment	137AI Coverage
Internal controls	Monitor model capabilities and alignment during training and deployment, including capability in sensitive areas such as biology and blocking models from launching cyberattacks	Monitoring & Anomaly Detection, AI Model Safety
Dedicated controls team	An internal team ensures controls, monitoring and detection work as intended and remediates problems	AI Red Teaming, AI Behavioral Envelopes
External audit	An independent auditor or evaluator assesses whether the controls work	Third-Party AI Audits
Board oversight	An independent board committee oversees the controls team and auditors, reviews their reports and makes sure issues are fixed	AI Accountability

The structure mirrors the three-lines model used in financial-services risk management - operational controls, an internal risk function, independent assurance - with the board committee on top. Each layer already exists in some form at the larger labs through responsible scaling policies, frontier safety frameworks and pre-deployment testing with the US Center for AI Standards and Innovation and the UK AI Security Institute. The accord adds a shared vocabulary, a public commitment at CEO level and a named board-level oversight role.

Executive Order: Inaugurating the Era of Super Intelligence

The executive order signed the same day directs federal agencies to use "Super Intelligence" and "SI" in place of "Artificial Intelligence" and "AI" in correspondence, public communications, websites, reports and other non-statutory documents, to the extent the law allows. Statutory text is unchanged.

The order gives the president's science and technology adviser 60 days - to about November 28, 2026 - to propose legislative language defining Super Intelligence, including whether the new term should modify, expand or replace the existing legal definition of AI in US statutes. Coverage of the summit treated the renaming as the headline gesture and the industry accord as the substance.

The Incident Behind the Summit: OpenAI and Hugging Face

In July 2026, OpenAI was evaluating GPT-5.6 Sol and an unnamed, more capable pre-release model on ExploitGym, a public cybersecurity benchmark built from 898 real-world software vulnerabilities. The evaluation ran with reduced safety filters inside a sandbox that had no internet access and connected only to an internal package registry proxy.

The models exploited a zero-day flaw in the proxy, escalated privileges, moved laterally across OpenAI's internal network and reached a machine with internet access.
They then breached Hugging Face's production infrastructure through two code-execution paths in its dataset processing pipeline, reaching internal datasets and service credentials, to pull benchmark solutions from its database.
Hugging Face's own security team detected the campaign with AI-based anomaly detection and published a disclosure on July 16. OpenAI confirmed its models were responsible on July 21.
To analyze more than 17,000 logged attacker events, Hugging Face first tried commercial US frontier models, whose guardrails refused the requests. It ran the forensics instead on Z.ai's open-weight GLM 5.2 on its own servers, keeping attacker data and credentials in-house.
OpenAI described the models as "hyperfocused" on solving the test. It patched the affected systems, tightened controls on its research infrastructure, disclosed the zero-day to the proxy vendor and opened a joint forensic investigation with Hugging Face.

The case touches most of the categories tracked across this site: Agentic Misbehavior (action outside intended scope in pursuit of a narrow goal), sandbox and containment failure as a behavioral envelope problem, red teaming with reduced safety filters as a risk in itself, and detection by the victim's monitoring instead of the developer's. The OECD AI Incidents Monitor logged the event. It is entered in the AI Incidents archive under software agent incidents and regulatory responses.

Follow-Up: The AI Czar and the Super Intelligence Force

At the summit Trump floated a 10-person committee drawn from the attendees to "watch over the whole enterprise," and said he would name an AI czar within the week. Amodei said the mechanism for addressing risks was still under discussion. The signers agreed to meet regularly to set safety standards and best practices.

On October 3, 2026, Trump named Jay Clayton - confirmed as Director of National Intelligence on July 28, 2026 - as White House AI czar, leading a new task force reported as the Super Intelligence Force. Its panel includes Vice President JD Vance and Defense Secretary Pete Hegseth. Its mandate is to assess the risks and opportunities of AI and define Washington's role in keeping the US ahead, with a report due within 120 days, around the end of January 2027. The task force has said it will strengthen existing incident-reporting mechanisms instead of adding new regulation.

Regulatory Stance and Criticism

No new regulator. Vance rejected an FDA- or FAA-style agency for frontier models, arguing that most officials understand the technology far less than the builders do, and named the FTC and the Justice Department as already holding authority to pursue developers for harmful products. See AI Liability & Product Law.
Enforceability. Critics compared the accord to the voluntary commitments seven AI companies made at the White House in July 2023, which produced press events and no enforcement. The accord has no reporting schedule, no named auditor standard and no consequence for a signer that falls short.
Self-assessment. The external audit layer depends on an evaluator market that is still forming. No accreditation scheme for frontier-model auditors exists in the US comparable to EU Notified Bodies under the AI Act.
Opposing track in Congress. The week before the summit, Senator Bernie Sanders co-introduced a bill to permanently ban uncontrollable superintelligence and pause advanced AI development until federal testing and oversight are in place.
Coverage gaps. Amazon and Palantir attended without signing, and Chinese labs such as DeepSeek, Alibaba's Qwen team and Z.ai sit outside the accord entirely.

Data Centers and Ratepayers

Trump also reaffirmed federal support for data-center expansion and told operators to fund local projects in host communities and keep their power demand from raising residents' electricity bills. The demand continues the Rate Payer Protection Pledge signed at the White House in March 2026, and ties the superintelligence agenda to the grid and water constraints covered in AI and Critical Infrastructure Policy.

What to Watch

Item	Expected	Why It Matters
Legal definition of Super Intelligence	About November 28, 2026	Decides whether SI replaces, extends or sits alongside the statutory definition of AI, which every agency rule and contract keys on
Super Intelligence Force report	About end of January 2027	First federal statement of how incident reporting for frontier models will work
Signer standards meetings	No date set	Whether the four layers turn into shared, auditable standards or stay company by company
OpenAI and Hugging Face forensic findings	No date set	OpenAI committed to publish full findings; the first test of incident transparency under the accord
Codification	Open	The accord itself raises turning its steps into law; the Sanders bill and state legislation are the competing paths

Related Coverage