A philosopher’s peer-reviewed paper, published the same week the U.S. government ordered a global AI shutdown, argues that the answer to AI risk is not restriction – it’s more innovation, done responsibly.
On the evening of June 12, 2026, Anthropic received a letter that no AI company had ever received before. Commerce Secretary Howard W. Lutnick directed the company to immediately suspend access to its two most powerful AI models — Claude Fable 5 and Claude Mythos 5 — for any foreign national, anywhere in the world.
Because Anthropic serves hundreds of millions of users globally and has no real-time way to verify citizenship, the only legally compliant response was a complete global shutdown of both models. Within hours, one of the world’s most widely used AI systems went dark for the majority of its users — not because of a technical failure, but because of a government order.
It was the first time in history that the U.S. government had used export control law to halt a commercial AI model already in broad public deployment.
The question the Mythos ban forced into the open was one that AI researchers, ethicists, and policymakers had been debating in theory for years: when a powerful AI system appears to pose a risk, should society slow it down — or push innovation forward to solve the problems innovation created?
Nine days later, a philosopher published a peer-reviewed answer.
A Third Way Between Prohibition and Recklessness
James Brusseau, a philosopher at Pace University in New York and the University of Trento in Italy, published a paper in the journal Law, Ethics and Technology on June 21, 2026, introducing what he calls “acceleration AI ethics” — a framework designed specifically for moments like the Mythos ban.
The paper argues that the debate around AI safety has been falsely framed as a binary choice: either prohibit dangerous AI systems, or step aside and let them advance unchecked. Brusseau’s framework proposes a third path — one where the risks created by AI innovation are systematically met with further innovation, rather than with restrictions that slow development or with indifference that ignores harm.
“The Mythos episode is the kind of dilemma this paper addresses,” Brusseau wrote.
The core argument is counterintuitive but grounded in a specific philosophy of how technological risks actually get resolved in practice. Rather than treating safety as a ceiling that limits how fast AI can advance, acceleration AI ethics treats safety as an engine that drives the next round of development. The risks themselves become the innovation brief.
What Actually Happened With the Anthropic Shutdown
To understand why Brusseau’s framework matters, it helps to understand the full sequence of events surrounding the Mythos ban.
The U.S. government’s directive on June 12 targeted Claude Fable 5 and Claude Mythos 5 — Anthropic’s most capable publicly accessible models at the time. The stated concern involved a classifier bypass vulnerability in Fable 5 that had been flagged as a national security risk, combined with broader export control concerns about frontier AI capabilities being accessed by foreign nationals.
The shutdown was total and immediate. Anthropic had no technical mechanism to screen users by citizenship in real time, so a full global suspension was the only compliant path.
Within two weeks, a partial restoration began. On June 26, 2026, Commerce Secretary Lutnick issued a revised directive that lifted certain restrictions on Mythos 5 specifically — making it available to U.S. government civilian agencies, national laboratories, critical infrastructure operators, and more than one hundred institutions named in a classified annex. Fable 5 remained suspended as of late June because the classifier bypass vulnerability that triggered the original order had not yet been technically resolved.
The structure of that restoration is itself significant. The government did not simply turn the models back on — it built an institutional access framework, tiered by organizational type and security clearance. It established, for the first time, that the U.S. government has both the will and the mechanism to selectively shut down and restore commercial AI models based on risk assessments.
How Acceleration AI Ethics Works in Practice
Brusseau’s paper does not remain in the abstract. To demonstrate how acceleration AI ethics functions as a real operational framework — not just a philosophy — he examines a concrete case study: the development of Telus’s GenAI customer support tool.
Telus, a major Canadian telecommunications company, built a generative AI system to handle customer service interactions across its full range of products and services. The development process immediately surfaced familiar concerns: hallucinations, prompt injection vulnerabilities, and privacy risks associated with handling sensitive customer data.
Under a precautionary approach, these concerns might have triggered delays, restrictions, or the abandonment of the project pending clearer regulatory guidance. Under acceleration AI ethics, they became something different — they became design requirements.
Brusseau describes how Telus’s engineers treated each identified risk as an innovation incentive. The response to prompt hacking vulnerabilities, for example, was not to restrict the system’s capabilities — it was to build a second AI agent whose specific function was to adversarially probe the first one, attempting to trigger harmful outputs so they could be flagged for human review and technical correction.
In other words: AI was used to challenge AI. The safety mechanism was itself an innovation, produced by the same forces that generated the risk.
This is the acceleration model in action. Each round of development generates new risks. Those risks define the objectives for the next round of development. Safety and innovation move forward together rather than trading off against each other.
The Values Underneath the Framework
Brusseau’s paper identifies several foundational values that enable acceleration AI ethics to function as a coherent framework rather than a rationalization for moving fast and ignoring consequences.
Decentralized governance is one of the most important. Acceleration AI ethics is skeptical of top-down prohibition models — not because rules don’t matter, but because centralized restriction tends to freeze development at a particular moment in time, preventing the innovation that might actually resolve the risk being regulated. Instead, the framework favors governance structures that emerge from the broad community of users and developers engaging with AI systems in practice.
Integration of ethics and engineering is another core value. Brusseau argues that the acceleration model only works when ethicists and engineers are collaborating from the earliest stages of development — not when ethics is applied as a compliance review after the technical work is done. Human concerns need to be converted into design challenges before the first line of code is written, not after the product is ready to ship.
Transparency about risk is the third pillar. Acceleration AI ethics does not claim that AI innovation is without danger. It claims that the most effective way to manage that danger is through further innovation guided by clear-eyed assessment of what the risks actually are — which requires the kind of honest, public accounting of AI vulnerabilities that the Mythos ban forced into the open.
Why the Timing Matters
The fact that this paper was published nine days after the Mythos shutdown is not a coincidence — it reflects a research trajectory that was already tracking the collision between AI capability development and governance frameworks.
The Mythos ban compressed that collision into a single news event. For the first time, a general audience could see concretely what it looks like when a government decides that an AI system poses sufficient risk to warrant immediate, global suspension. The abstract policy debate about AI safety became a real-world event with immediate economic and operational consequences for millions of users and thousands of businesses.
Brusseau’s acceleration AI ethics framework is positioned as a response to exactly that kind of moment — providing a principled way to think about what should come next, rather than cycling between the two unsatisfying options of permanent prohibition or unmanaged resumption.
What This Means for the AI Industry
The Mythos ban and Brusseau’s paper together signal something important for companies building or deploying AI tools: the governance landscape has structurally changed.
The restoration framework that Commerce Secretary Lutnick implemented on June 26 — tiered access by entity type, classified annexes, partial reinstatement based on technical remediation — is not a one-time exception. It is a template. It demonstrates that the U.S. government has now built and tested a mechanism for graduated AI model access control, and is prepared to use it again.
For enterprise technology leaders, security teams, and AI developers, the operational implication is straightforward: AI governance can no longer be treated as a future regulatory question. The Fable-Mythos episode established that government-directed model suspension is a live risk that needs to be factored into AI deployment planning right now — alongside questions of workforce access, export control compliance, and contingency planning for sudden service interruption.
Brusseau’s framework offers one way to respond to that reality — not by slowing down, but by building safety into the innovation process itself, making each round of development more robust than the last.
The Broader Debate Continues
Acceleration AI ethics is not without critics. Skeptics of the framework argue that treating risk as an innovation incentive places too much faith in the ability of developers to identify and resolve their own blind spots — and that some categories of AI risk may require external constraints that no amount of internal innovation can substitute for.
The Mythos case itself is instructive on this point. The classifier bypass vulnerability that triggered the original government directive was not caught and fixed by Anthropic’s internal development process before it became a national security concern. It required external intervention — a government order — to force a halt.
Whether that represents a failure of internal governance that acceleration AI ethics could have prevented, or evidence that some risks genuinely require external regulatory action regardless of how sophisticated the internal framework is, remains an open and important question.
What is clear is that the question is no longer theoretical. The government has now demonstrated it will act. The industry now needs frameworks for deciding how to respond — and Brusseau’s paper is one serious attempt to provide one.
“Acceleration AI Ethics and the Telus GenAI Conversational Agent” by James Brusseau was published in Law, Ethics and Technology (ELSP) on June 21, 2026. Brusseau is a professor in the Philosophy Department at Pace University, New York, and the Department of Information Engineering and Computer Science at the University of Trento, Italy.
