Washington gave its new AI task force 120 days to prove it can hear trouble

THE SIGNAL IN ONE SENTENCE

A task force is government shorthand for: several agencies own one problem, nobody owns all of it and somebody has finally admitted the handoffs matter. President Donald Trump announced a new federal artificial intelligence task force on October 4 and named Director of National Intelligence Jay Clayton as its chair. The administration calls it the Super Intelligence Force, following a September 29 executive order that told executive agencies to replace the words artificial intelligence and AI with super intelligence and SI in non-statutory government communication. The vocabulary is flashy. The assignment is more useful. Reuters reports that the group has 120 days to assess the risks and opportunities of AI, recommend what role the federal government should play and examine how the government receives and responds to reports of breaches, hacks and other AI incidents. The group is expected to report to the president and White House Chief of Staff Susie Wiles. Clayton will be joined in leadership by Federal Trade Commission Chair Andrew Ferguson, Pentagon technology chief Emil Michael and Office of Personnel Management Director Scott Kupor. That roster puts four very different government muscles in one room: national intelligence, consumer protection and competition, defense technology, and the civilian workforce. The plain signal is not that Washington has solved AI governance. It has started a 120-day coordination exercise without yet publishing the public machinery that would let people judge it. The announcement did not arrive with a standalone charter on the reviewed White House or agency pages, a dedicated public website, a budget, staffing details, a meeting calendar, a public comment process, an incident-reporting form, a list of deliverables or a precise rule for when the 120-day clock starts. Reuters and other reporting describe a charter and its mandate, but the charter itself was not available in the reviewed official sources. That distinction matters because a task force can become one of two things. It can be a routing layer. A researcher, company, hospital, school, infrastructure operator or ordinary user reports a serious AI failure once. The government knows which agency should receive it, what evidence to preserve, who can act under existing law and what the public will eventually learn. Or it can become a conference table with excellent stationery. The membership hints at the first version. The DNI can see national-security threats and coordinate across the intelligence community. The FTC can investigate deception, unfair practices, consumer injury and competition concerns. The Pentagon has technical experience buying and testing frontier systems for high-stakes environments. OPM can translate policy into training, hiring and operating rules for the federal workforce. Those responsibilities also create the main governance question: who speaks for the person affected by the system? The announced leadership is entirely federal and heavily weighted toward security, industry and administration. Consumer groups, workers, researchers, state officials, civil-rights organizations, educators, health professionals and people affected by automated decisions may be consulted, but they do not appear to hold named leadership seats in the announcement. Consultation is not the same as authority. A listening session can gather stories. It cannot guarantee that those stories survive the trip into a recommendation. The task force also enters a federal landscape that is not empty. A June executive order already directed agencies to create an AI cybersecurity clearinghouse, develop classified benchmarks for advanced cyber capabilities and design a voluntary early-access framework for covered frontier models. The September naming order requires the White House science adviser to propose a federal definition of super intelligence within 60 days. The FTC has its own consumer-protection and competition powers. Defense and intelligence agencies already operate model-testing and security programs. The new group therefore does not need to invent another map of the map. It needs to show where existing systems fail to connect. Imagine that a model used by a water utility begins taking actions outside its approved task. The operator sees an unusual command, the vendor sees a model trace, a cloud provider sees network traffic and a security researcher finds a reproducible trigger. Four parties hold four pieces of the incident. None may know whether to call CISA, the FBI, the FTC, a sector regulator, a state attorney general or the model developer. A useful federal system would give them one intake route with a clear severity ladder. It would define what to report, how quickly to report it, how sensitive logs are protected, when the vendor must be notified, when another agency takes the lead and when the public receives a sanitized account. The hardest part is not the web form. It is deciding what counts as an incident. A jailbreak that produces an ugly answer is not the same as a model that exfiltrates credentials. A hallucinated medical claim is not the same as an autonomous system changing a patient's record. A benchmark result is not a breach. A security research demonstration is not automatically a crime. A misuse report can implicate user privacy, free expression and due process at the same time that it raises a genuine safety concern. The task force needs a taxonomy sturdy enough to separate those cases without pretending the boundaries are easy. It also needs escalation rules. A low-severity report may belong in a vendor's bug tracker. A repeated consumer deception pattern may belong at the FTC. A vulnerability in critical infrastructure may require CISA and the sector regulator. A national-security intrusion may move into classified channels. A civil-rights complaint may need a completely different legal path. If every report becomes secret, the public cannot learn from failure. If every report becomes public immediately, investigators may expose victims, methods or live vulnerabilities. The adult answer is tiered disclosure with deadlines, not one rule for every case. The 120-day report should therefore be judged by a short list of concrete artifacts. First, publish the charter and membership rules. Readers should know the group's authority, conflicts, voting structure, consultation process and relationship to existing agencies. Second, publish an incident taxonomy. Define severity, evidence requirements, responsible agencies, response times and appeal routes in plain language. Third, create one public intake door with protected channels for sensitive material. Nobody should need an org chart and a law degree to report a serious failure. Fourth, disclose aggregate data. How many reports arrived, what types they covered, how many were referred, how long triage took and what corrective actions followed? Fifth, separate recommendations from enforceable authority. A task force can advise. Congress can legislate. Agencies can act only within the powers the law gives them. A recommendation should name who can implement it and what legal step remains. Sixth, publish minority views. If experts disagree about acceptable risk, liability, open models or reporting thresholds, burying the disagreement does not make it disappear. It only makes the final document look more certain than the evidence. This is also where the administration's new name becomes a practical nuisance. The September 29 order says super intelligence means the technologies already covered by the statutory definition of artificial intelligence unless law or later action changes it. That means the new phrase does not itself create a new technical category. It covers ordinary AI systems as well as frontier ones in executive communication. Names can shape policy. They can also blur it. Calling everything super intelligence may make a spreadsheet classifier sound like a world-changing machine. It may encourage agencies to treat very different systems as one political object. The task force's first useful act would be to classify systems by capability, access, autonomy, consequence and deployment context instead of relying on the grandeur of the label. The composition of the group creates another tension. Putting the intelligence director in the chair signals that the administration sees AI partly as a strategic race. That may help coordinate serious cyber and national-security risks. It may also pull attention toward adversaries and away from mundane domestic harms such as a wrong benefits decision, an inaccessible hiring filter or a deceptive subscription chatbot. Ferguson's presence gives consumer protection a seat. Kupor's role can connect policy to federal workers. Michael's role can bring technical procurement and testing experience. The quality of the work will depend on whether those lanes remain visible or collapse into one question about national advantage. Winning a technology race and protecting people are not automatically opposites. They are not automatically the same thing either. The public should resist two easy reactions. One is to treat the task force as proof that the government now has AI safety handled. It does not. A body with a deadline is an input, not an outcome. The other is to dismiss the whole exercise because task forces often produce reports. Reports can matter when they define a durable intake process, expose gaps between agencies, assign accountable owners and create measurable follow-up dates. The most revealing page may be the appendix. Look for named agencies, legal authorities, response clocks, evidence standards, unresolved disputes and public metrics. Look for the cases that did not fit neatly. Look for commitments that still exist six months later. The new federal AI task force has 120 days to turn a dramatic name into ordinary machinery. That is exactly the kind of machinery government needs: a place where somebody can report trouble, somebody knows who owns the next step and the public can eventually see whether anything happened. The clock is not a strategy. It is a promise to deliver one.