Trump’s latest AI czar has already resigned

The Revolving Door at CAISI: What Three Departures in Four Months Mean for AI Standards

Chris Fall has resigned as director of the Center for AI Standards and Innovation (CAISI), the agency confirmed to multiple news outlets. He was appointed just three months ago. Fall is the third person to leave the leadership role since March, following venture capitalist David Sacks (who stepped down as White House AI and crypto czar) and Collin Burns, who reportedly was pushed out in less than a week because of his prior work at Anthropic. The Trump administration provided no reason for Fall’s departure.

For developers and teams building on frontier models, this leadership churn at the primary U.S. body for AI technical standards creates a vacuum of predictability just as regulatory and evaluation frameworks are being contested from multiple directions.

CAISI’s Mandate vs. Its Instability

CAISI operates under the National Institute of Standards and Technology (NIST). Its formal mission is to develop technical standards and testing methods for AI models, as well as to assess cybersecurity risks. It is the principal U.S. organization tasked with creating the infrastructure for evaluating model safety, robustness, and reliability — the very sorts of benchmarks that developers rely on when choosing which model to integrate into a product or pipeline.

Yet three directors in roughly four months suggests an agency that cannot execute a sustained technical agenda. Standards work, by its nature, requires continuity: building consensus across labs, running longitudinal evaluations, and publishing reproducible methodologies. A revolving door at the top makes it difficult to maintain relationships with industry partners, set long-term testing priorities, or respond to urgent model incidents with a coherent playbook.

Fall came with relevant credentials: he previously directed the Department of Energy’s Office of Science during the first Trump administration, served as acting director of the DOE’s Advanced Research Projects Agency-Energy, and worked in the Office of Naval Research. That background in large-scale government R&D might have been valuable for CAISI’s technical work, but the short tenure erodes any benefit.

What the Recent Model-Risk Brouhaha Reveals

The absence of CAISI from the most high-profile model-risk event of the year is telling. In June, the U.S. Commerce Department invoked an obscure export control directive that effectively forced Anthropic to pull its Mythos and Fable models from the market. The ban was lifted by the end of the month when Commerce Secretary Howard Lutnick expressed satisfaction with Anthropic’s safety plans. CAISI — the agency built to assess such risks — was not the lead actor.

Then, earlier this month, the White House signed an executive order creating a new AI safety oversight program called “Gold Eagle,” a clearinghouse for cybersecurity vulnerability coordination that names the Commerce Department and the Department of Homeland Security. As CNBC reported, CAISI was not among the federal organizations mentioned in the program.

This marginalization of the agency supposedly tasked with AI standards is significant for developers. When a crisis like the Anthropic ban occurs, the community needs transparent, technical evaluation criteria — not behind-closed-doors executive action. Without a stable, authoritative body to provide those criteria, model providers and downstream builders face regulatory whiplash. A model that passes one set of internal safety checks today could be pulled tomorrow based on a directive that references no public standard.

The Industry Alternative Gains Momentum

Google DeepMind CEO Demis Hassabis has begun calling for the creation of an independent, industry-run standards body to regulate frontier AI, modeled after FINRA — the self-regulatory organization for securities markets. That is essentially the same mission CAISI was formed to tackle. The fact that a prominent lab leader is advocating for an industry alternative suggests deep dissatisfaction with the government’s approach.

For developers, this signals a potential fork in the road. If CAISI remains unstable and sidelined, de facto standards may emerge from industry coalitions or from the most powerful labs themselves. That could lead to fragmentation: one evaluation regime from a Google-led body, another from a consortium of open-weight model advocates, and yet another from the Commerce Department acting ad hoc. Builders who need to certify their applications across multiple models will face rising compliance complexity. In that environment, tools like AI coding agents that abstract away model selection become more valuable — but only if they can track and adapt to shifting standards. Our comparison of the leading AI coding agents in 2026 highlights how different agents handle model compatibility and safety checks, which will become critical as certification regimes proliferate.

Chinese Open Models and the Testing Transparency Gap

CAISI’s work on Chinese open-weight models highlights another tension. The agency has released a few reports on the capabilities of open-weight models such as Z.ai’s GLM-5.2 and DeepSeek V4 Pro. (Open-weight means the model can be publicly downloaded and run locally, but training code and datasets are not available.) Yet CAISI hasn’t discussed much about its processes for testing these models. Since July 9, TechCrunch has sent multiple inquiries to both the Department of Commerce and NIST about how its LLM evaluations work and has received no response.

This transparency gap matters directly to developers using open-weight models. If a government body issues a report claiming a model has certain safety or capability characteristics, but does not explain its methodology, it becomes impossible for the community to validate or challenge those findings. Developers integrating a model like DeepSeek V4 Pro into a product have no way to know whether CAISI’s assessment is rigorous or politically motivated.

The timing is especially delicate. This weekend saw handwringing over Chinese AI lab Moonshot’s new version of its open model Kimi, which performed competitively against flagship frontier models. The administration was reportedly weighing efforts to ban Chinese open models — a move that prompted immediate debate. David Sacks, the former CAISI lead, argued publicly that regulations should not be used as a protectionism strategy for U.S. proprietary AI labs.

For developers, the prospect of a ban on open-weight models from entire countries raises the stakes of the standards discussion. Many builders rely on open models for fine-tuning, cost efficiency, and transparency. A ban — or even the threat of one — could force teams to redesign their stacks around proprietary APIs. The lack of a stable, transparent standard-setting body means the rules could change without technical justification or public input.

Practical Takeaways for AI Builders

First, do not assume CAISI will be the definitive source of AI safety standards. The revolving door suggests the agency lacks the institutional stability to lead on long-term evaluation work. Developers should watch for standards emerging from other bodies — whether industry consortia, international groups, or other federal agencies like the Department of Homeland Security under the Gold Eagle program.

Second, build modular evaluation pipelines. If you’re using an AI coding assistant or integrating a frontier model, design your safety and verification layer to be swappable. An evaluation method that depends on a single government-mandated checklist may become obsolete or contradicted. Our comparison of Cursor vs. Claude Code shows how different development agents handle model provenance and safety checks — a capability that will become increasingly important as evaluation regimes diverge.

Third, engage with the transparency conversation. CAISI has not answered basic questions about its LLM evaluation methodology. Developers who rely on open-weight models from any source — Chinese, American, European — should push for reproducible, published testing standards. If the official body won’t provide them, industry-led initiatives (like the one Hassabis is proposing) may be the only credible path forward.

Fourth, monitor the regulation-vs-protectionism debate. The discussion around banning Chinese open models is not purely technical. It has political and trade dimensions that could affect model availability regardless of safety. Builders should be prepared for regional model availability shifts, and consider maintaining fallback options across different model families and jurisdictions.

The Bottom Line

Chris Fall’s resignation is not just a personnel story. It is a symptom of an agency that has been unable to establish itself as a credible, stable center of gravity for AI standards. Between the rapid succession of directors, the sidelining from major model-risk actions, and the lack of transparency about testing methodologies, CAISI is currently offering little of the certainty that developers need to make long-term technical decisions.

The industry is responding by floating alternatives. Whether those alternatives gel into a coherent new framework or create further fragmentation remains to be seen. What is clear is that builders cannot wait for clarity from a revolving door. They need to invest in flexible evaluation architectures and stay engaged in the standards conversation — because the rules of the road for AI have not been written yet, and the pen is being dropped with increasing frequency.

Source: TechCrunch. Details as reported; verify specifics at the source.