On Friday, June 12, Anthropic abruptly disabled its two most capable AI models, Claude Fable 5 and Mythos 5, for every customer worldwide. The models had launched only three days earlier. The cause was not a server crash or a billing dispute: the company says it received a U.S. government export-control directive ordering it to cut off access for any foreign national, inside or outside the United States. Unable to screen users one by one in real time, Anthropic shut the models down for everyone. A firm that had wired those models into its workflow on Wednesday found them gone on Friday — through no fault, no warning, and no decision of its own.
It is worth sitting with the detail that makes this unsettling for lawyers: the tool did not fail. It worked exactly as designed, right up until a third party with authority over the vendor decided it should stop. No business continuity plan built around uptime, redundancy, or vendor reliability would have caught this, because the point of failure was none of those things.
And it does not take a government order to put a firm in that position. Consider two more ordinary mornings.
It is 7:15 on a Monday morning. A trial lawyer is sitting in the courthouse parking lot, engine running, panel list in hand. Jury selection starts in ninety minutes. For the past three weeks, her firm's AI platform has been building profiles on every potential juror in the pool — employment history, social media patterns, prior jury service, the kind of granular picture that used to take a team of consultants and twice the budget. This morning, the platform is returning a single error message. No estimated resolution time. She has the names. She has nothing else.
At the same moment, in a conference room in Chicago, a transactional lawyer is staring at a due diligence summary that is suddenly incomplete. Closing is in seventy-two hours. His team has spent the last three weeks running forty thousand documents through an AI review platform that was, until last night, flagging risk provisions, surfacing inconsistencies, and building the representation and warranty analysis the deal depends on. The platform is down. The seller is not extending. The documents are still in the data room, unreviewed.
Neither of these lawyers has a business continuity plan for this moment. Neither of them thought they needed one. Law firms have business continuity plans for power outages, for the illness of key personnel, for data breaches. Almost none have a documented plan for the unavailability of an AI tool they have come to depend on. That gap is going to cost someone a client, a deadline, or worse. It may already have.
Four Ways AI Goes Away
There are at least four distinct scenarios worth planning for, and they are not equally recoverable.
The first is the outage — temporary, unplanned, usually resolved within hours. Every major AI platform has experienced them, and as these tools become more deeply embedded in legal workflows, the outage that hits at the wrong moment becomes a real professional risk. A firm that treats AI as a convenience recovers quickly. A firm that has rebuilt its core workflows around a single platform may not.
The second is the pricing change — structural, potentially permanent, and more disruptive over time than any outage. AI tools are currently priced aggressively to gain market share. That strategy has a shelf life. When pricing normalizes — or when a vendor decides that law firms will pay significantly more than individual users — the economics of AI-dependent workflows can change overnight. We have already seen meaningful price increases from major providers, and the direction of travel is not toward cheaper. A firm that built its efficiency model around today’s pricing may find that model no longer works at tomorrow’s rates.
The third is discontinuation or acquisition — the most permanent and the least discussed. The legal AI market is crowded with well-funded startups, and market consolidation is coming. When a legal AI tool gets acquired, the acquirer’s priorities are not necessarily the same as the acquired company’s customers. Terms change. Features disappear. Tools get folded into products that serve different markets. Firms that built customized workflows around a platform that no longer exists in its prior form are not recovering their investment.
The fourth is the regulatory or government-ordered shutdown — the scenario almost no one was planning for until the Fable 5 suspension put it on the front page. This is the failure mode that does not care how reliable your vendor is, how much redundancy you bought, or how well you negotiated your contract. A government acting on national-security, export-control, or sanctions authority can require a vendor to cut off access overnight, and the vendor will comply. The trigger sits entirely outside the customer relationship. It can reach a tool that is working perfectly, profitable for the vendor, and central to your practice, and it can do so with no notice and no appeal. For firms that handle cross-border matters, represent foreign-owned clients, or employ attorneys who are not U.S. citizens, this is not a remote contingency — it is a live one.
Why the Risk Is Larger Than It Looks
The severity of this risk depends almost entirely on which level of AI use your firm has adopted — a framework we described in detail in our earlier field guide to AI in legal practice.
At the Foundation level, the risk is real but manageable. If the AI tool that drafts your first-cut correspondence goes offline, you draft it the old way. The inconvenience is meaningful; the damage is limited. Foundation-level AI use is largely interchangeable — one general-purpose platform can often substitute for another with modest adjustment.
At the Advanced and Expert levels, the exposure is fundamentally different. If your litigation preparation is built around tools that research judges, model jury profiles, compile expert witness histories, and generate custom trial analytics — and those tools sit on a third-party platform whose terms or availability you do not control — an outage is not an inconvenience. It is a capability gap that cannot be closed by switching to a different chatbot.
The firms most exposed are the ones that have done exactly what they were told to do: invested seriously in AI, built sophisticated workflows, and integrated those workflows into their practice. The investment is real. The vulnerability is the same.
What Business Continuity Actually Requires
Most law firm business continuity planning treats technology as infrastructure: keep the servers running, back up the data, have a remote access protocol. AI tools fit uneasily into that framework because the risk is not data loss — it is capability loss. You still have the files. You have lost the ability to process them the way you built your practice to process them.
A genuine business continuity plan for an AI-dependent practice starts with an honest inventory: which workflows depend on which tools, and what happens to each workflow if the tool is unavailable for a day, a week, or permanently? That inventory will identify single points of failure — places where one tool’s unavailability stops a critical process entirely. Those are the places that need redundancy.
Redundancy in an AI practice does not mean having a backup subscription to a second platform, though that helps for Foundation-level tools. At higher levels, it means ensuring that the institutional knowledge embedded in your AI workflows is also documented as a process that can be executed, more slowly and more expensively, without the AI. If the only people who understand how a workflow runs are the AI system and the partner who built it, the workflow is fragile in two different ways.
It also means maintaining the underlying legal skills that AI has augmented. A litigation team that has delegated judicial research entirely to an AI tool, with no lawyers who remember how to do it the traditional way, has a readiness problem that extends well beyond any single outage. AI should make your lawyers faster and better, not replace the judgment and capability that makes them lawyers.
The Planning Question Worth Asking Now
The firms that will handle AI disruption best are the ones that asked a simple question before the disruption happened: if this tool were unavailable tomorrow, what would we do?
For Foundation-level tools, the answer is usually straightforward — adjust, substitute, slow down temporarily. For tools that have been built into the core of a practice, the answer requires more thought. And for firms that have built on third-party platforms they do not control, the answer to that question may reveal a strategic vulnerability that no amount of good work product can fix after the fact.
The Fable 5 shutdown will likely be resolved — Anthropic has called the suspension temporary, and access to its other models was never affected. But the lesson outlasts the event. The firms that came through the week unbothered were not the ones with the best vendor; they were the ones who had already answered the question of what they would do without it. Asking that question in advance is the whole point of a continuity plan.
The next post in this series takes up the related question of platform dependency: why the competitive advantage you rent is not really yours, and what it means to own your methodology rather than lease it from someone else. And in a companion piece following close behind, "Money for Nothing," we turn from the tools to the people who use them — asking why firms are paying a record $235,000 for a first-year associate at the very moment AI is automating the work that salary was meant to fund, and who ends up paying for the contradiction.
