The practical effect: Anthropic is excluded from government contracts. The symbolic effect: if the government is willing to do it to Anthropic, perhaps they are willing to do it to anyone.
It’s a shocking move. That will resonate beyond its legal effects.
But I’m at 91% that Anthropic will still be de facto blocked at that time.
Here’s my Monte Carlo model and the historical data that informs it[2].
What “Supply Chain Risk” Actually Means
The relevant statutes — 10 U.S.C. § 3252 and 41 U.S.C. § 4713 — give the Secretary of Defense authority to exclude vendors from contracts when they pose unacceptable supply chain risk.
10 U.S.C. § 3252 focuses on the DoD, 41 U.S.C. § 4713 affects federal agencies more broadl3[3]y. 4713 is more vague on how it is meant to apply, but there is a chilling effect, which is why Anthropic is taking this to court. Federal agencies will stop doing business regardless. Private businesses that serve Federal agencies may also. The designation is immediate.
But that’s not all. President Trump truthed that all agencies should immediately cease using Anthropic’s technology. This too has real world effects. Agency heads are appointed by President Trump and will not easily contradict him. Many private companies wish to signal cooperation with this administration. And these effects will very likely not be overturned by an Anthropic legal victory. I’ll cover these at the end.
So what might things look like for the two statutes on May 1st?
On in the Northern District of California, challenging the entire exclusion — the §3252 designation, the Presidential Directive, and all agency actions taken in response — and seeking a TRO and preliminary injunction that would block them
The legal arguments are strong. Anthropic’s core claim is that the Pentagon stretched the supply chain risk statute beyond its statutory authority — it was designed for procurement decisions, not as a general-purpose tool to punish a domestic company for its AI ethics policies. There are also First Amendment dimensions: the designation came shortly after Anthropic publicly refused to remove restrictions on mass surveillance and autonomous weapons use. Legal commentary has been blunt — Lawfare called the designation something that “won’t survive first contact with the legal system.”
Base rates
The closest precedent comes from the 2021 wave of “Chinese Military Company” designations under EO 139595 and the 1260H that followed it[5]. The DoD designated dozens of companies with little public explanation. Several sued for Preliminary injunctions (PIs) to hold up the effects until the case was decided in court. All cases where the company sued that I can find details of are below. In the few cases that sought emergency relief, they got it and either won their cases or the Government withdrew.
Xiaomi — CMC list (EO 13959). Sought PI → granted day 39. DoD withdrew.
Luokung — CMC list (EO 13959). Sought PI → granted day 62. Not re-listed.
DJI — Section 1260H list. No PI sought → lost on merits after 18 months.
Hesai — Section 1260H list. No PI sought → lost on merits after 18 months.
AMEC — Section 1260H list. Sought PI → DoD voluntarily removed before ruling.
YMTC — Section 1260H list. Sought PI → pending (filed Dec 2025).
127 more that didn’t sue.
There are other cases, such as Huawei and Esquel, but those aren’t directly comparable. Huawei was banned by an act of Congress that named it specifically, not by an executive agency decision. Esquel challenged a Commerce Department listing, but the court found Commerce was using that tool exactly as intended. Generally, courts defer much more to Congress than to executive agencies on national security.
The judge
The N.D. Cal case drew Judge Rita F. Lin, a Biden appointee. In Thakur v. Trump, a case involving APA violations and First Amendment retaliation against government contractors, Judge Lin granted a preliminary injunction on a similar legal theory Anthropic is advancing. The 9th Circuit upheld her reasoning on appeal.
That doesn’t guarantee the same outcome. But it seems likely to me. The model puts P(N.D. Cal grants relief) at 83%.
The two-statute problem
Here’s the wrinkle that makes this harder than it looks. The cases are about different statutes, and they live in different courts.
§3252 (DoD-only procurement exclusion) is challenged in N.D. Cal. That’s the case before Judge Lin, and it’s moving fast — hearing scheduled March 24.
§4713 (government-wide exclusion) is challenged in the D.C. Circuit. By statute, challenges to these orders must be filed there. The D.C. Circuit is a separate court with its own timeline.
For the Metaculus question to resolve NO, both statutes need to be blocked. If the N.D. Cal injunction only covers §3252, then §4713 keeps Anthropic locked out of federal agencies. As I’ve mentioned above, this may be the case even if both are overturned, since President Trump has still make his view clear, but the Metaculus resolution criteria seem very reasonable to me.
The key question: could Judge Lin’s N.D. Cal injunction reach broadly enough to cover §4713 too?
Anthropic’s complaint names five counts, including ultra vires Presidential Directive and APA §558 challenges that sweep broadly. The prayer for relief asks the court to block “all Challenged Actions.” But “Challenged Actions” is defined as the Presidential Directive, the Secretarial Order, and actions taken in response — it does not explicitly mention the §4713/FASCSA designation.
I give the probability of the N.D. Cal injunction covering §4713 at 45%. Not impossible — the complaint’s broad framing could pull it in — but the complaint wasn’t drafted to target §4713, presumably deliberately, and Judge Lin might reasonably say that’s the D.C. Circuit’s job.
The D.C. Circuit is the backup. Model gives it a 70% chance of granting relief, but its timeline is slower and less certain. It just might not have judged by May 1st.
Timing
The N. D. Cal hearing is scheduled March 24th. The model expects a ruling within days of that hearing.
I give the D.C. ruling a median of March 30th, 21 days after filing.
The May 1 deadline is 53 days from filing. Xiaomi’s PI came at day 39. Luokung’s at day 62. Over 99% of simulations have at least one court ruling before May 1.
The stay risk
If a court blocks the designation, the government can seek an emergency stay while it appeals. This administration has been aggressive about stays. The model gives an 85% chance the government seeks one.
But getting a stay granted is hard. The combined probability through the 9th Circuit and potential SCOTUS escalation is about 22%[6]. Neither of the Xiaomi/Luokung blocks were successfully stayed. The government would need to show a “reasonable probability of success on appeal,” which is tough when the lower court just found your process arbitrary and capricious.
Path 2: The Government Withdraws (~8%)
The government could pull the designation at any time. In the model, this mostly happens after courts have ruled — either as a face-saving exit after losing (the Xiaomi pattern, where DoD called it a “mutual agreement” ~60 days after the PI) or as a quiet removal (the AMEC pattern, no announcement).
About 8% of simulations end with a government withdrawal. Most of these happen after court setbacks, not before. I think that Anthropic is quite minded to do a deal with the Government, as long as it doesn’t cross their red lines, but I expect most of the probability mass of that after May 1st. The model gives 6.6% unilateral withdrawal, 1.6% withdrawal via deal.
The Polymarket market on “Will Anthropic make a deal with the Pentagon?” prices a deal at about 14% before May 1st (at time of writing). My model’s deal path before May 1st is much smaller.[7] This hasn’t been the central focus of the model, but I’m loath to push it up based on vibes. Anthropic is currently blocked, the courts might rule in the Government’s favour, and there can be stays afterwards and appeals (likely after our deadline). I think unless Anthropic is going to cross their red lines (very unlikely) a deal comes if the Government faces setbacks, not in the next month.
Path 3: The Designation Stands (~43%)
This is the scenario the Metaculus community is roughly pricing in. My simulation agrees it’s substantial.
The biggest YES path isn’t “courts reject Anthropic.” It’s the two-statute structure.
~13% — N.D. Cal rules narrowly in Anthropic’s favour, no DC judgment, §4713 survives. Judge Lin blocks §3252, but her injunction doesn’t reach FASCSA. The D.C. Circuit hasn’t ruled yet, or rules against Anthropic. The government-wide exclusion stays in place. This is the single most likely YES outcome and it’s the one most forecasters seem to be missing.
~12% — Court blocked, but stay granted. A court issues the PI, the government gets an emergency stay reinstating the designation and this runs through May 1st.
~12% — N.D. Cal denies relief outright. Judge Lin rules in favour of the Government, likely on national security grounds. Not the most likely outcome given her track record, but not negligible.
~5% — Both courts deny relief. Everything goes wrong for Anthropic. Both courts find the designation lawful.
~1% — Limbo. A catch all set of weird outcomes. Neither court rules before May 1. Very unlikely given the March 24 hearing.
The Full Monte Carlo
What Metaculus “NO” Actually Means
One thing worth flagging: Metaculus resolving NO does not mean Anthropic gets its government business back. The supply chain risk designation is one of several mechanisms keeping Anthropic out. Even if the courts block it:
The Presidential Directive ordering agencies to “cease all use of Anthropic products” is a separate instrument. A narrow injunction might not touch it.
The administration could issue new restrictions under IEEPA or the Defense Production Act. I haven’t thought deeply about this, but I give that about 10% before May 1st.
No contracting officer wants to be the first to sign an Anthropic deal while the White House is actively hostile.
The model estimates a 91% chance Anthropic remains effectively blocked even if the designation is lifted.[8] The Metaculus question is a legal question, not a practical one. I can see this changing via a deal (likely after May 1st or the Government losing so decisively that agencies feel comfortable using Anthropic products again. I think these outcomes are very unlikely before the deadline. Note that since there is a 6 month wind-down period, Anthropic might still have Government business but it’s still tailing off.
Metaculus, I was wrong.
The Metaculus community is at 50% YES. My model produces 43%. This seems close enough that I’ll just go with mine.
The original version of this post had a model that returned 14% and contained the following pair of paragraphs:
This confuses me, but my current read is that when there isn’t an effortful comment, most of the community doesn’t think that hard about multi-step processes. There is currently a designation in force, will there be in 50 days, who knows? I doubt the median Metaculus user has researched how quickly these tend to be challenged in court.
My bad. Though I still do think this is a bit true. But I’ll defer a bit more in future.
In my defence my forecast was always a blend of my thoughts and Metaculus’ (it never got below 23%). But in this case, I was wrong. I thought that either court could rule on both determinations. Now that seems much less likely.
What I’m Watching
The March 24 hearing. Judge Lin’s line of questioning will signal how she’s reading the case.
The government’s response brief. In Xiaomi, the government’s inability to articulate a factual basis was fatal. I look forward to the DoD’s response.
Whether Judge Lin addresses §4713. If she signals willingness to reach the FASCSA designation, P(CA broad) goes up and YES probability drops.
D.C. Circuit scheduling. Any indication of expedited review there would move the model, making the Metaculus NO resolution more likely (though not changing the overall picture much).
Executive orders. I find President Trump hard to Predict. This model is primarily about the designations, but for a more accurate picture, I am watching his statements on this. There is more the Government can throw at it.
The Bottom Line
My model says 43% chance the designation is still in effect on May 1. And a 91% chance that in some deep sense Anthropic’s business with federal agencies is still tailing off.
Fifty days to find out.
Forecast as of March 12, 2026. Based on a 200,000-simulation Monte Carlo model.
CA ruling median 16→30 days — YES moves 42.6%→42.4% — 0.3pp swing
Model code available to paid subscribers on request, I intend to write more of these behind a paywall, currently I have a lot of analysis on whether there will be an Iran ceasefire and on silver and oil prices. Thanks Metaculus. This analysis took 2 - 5 hours, and the writing (which involved a lot more analysis) about 8 hours. I would charge somewhere between $1000 and $5000 for analyses like these.
I get $1000, if I beat the community prediction (hard) I get $3500. My hourly rate for this is currently about minimum wage, but I think it's mainly competing with Twitter.
I think. I have focused my core error correction on the model itself. I can imagine I get one statement like this wrong.
In my view, it is much better to get the overall picture right and the details wrong, than the other way around. I am forecasting the chance these laws are still in effect.
This is the kind of probability I haven't checked that carefully. The sensitivity analysis shows it isn't that important. Am I wrong? Disagree in the comments.
One might say that this is the lede and that this entire article is confusingly written in regard to that. One also might say that we. are about 7 hours into a 1 hour blog post, so bleh.
On March 4th, 2026, the Pentagon did something it had never done to an American company before: it formally designated Anthropic a supply chain risk under U.S. defense procurement law.
The practical effect: Anthropic is excluded from government contracts. The symbolic effect: if the government is willing to do it to Anthropic, perhaps they are willing to do it to anyone.
It’s a shocking move. That will resonate beyond its legal effects.
As part of the Metaculus Spring Tournament, where I am this season’s paid antagonist,[1] I’ve been researching the following question “Will Anthropic be a designated supply chain risk on May 1, 2026?” The Metaculus currently puts it at 50/50. I’m at 43%.
But I’m at 91% that Anthropic will still be de facto blocked at that time.
Here’s my Monte Carlo model and the historical data that informs it[2].
What “Supply Chain Risk” Actually Means
The relevant statutes — 10 U.S.C. § 3252 and 41 U.S.C. § 4713 — give the Secretary of Defense authority to exclude vendors from contracts when they pose unacceptable supply chain risk.
10 U.S.C. § 3252 focuses on the DoD, 41 U.S.C. § 4713 affects federal agencies more broadl3[3]y. 4713 is more vague on how it is meant to apply, but there is a chilling effect, which is why Anthropic is taking this to court. Federal agencies will stop doing business regardless. Private businesses that serve Federal agencies may also. The designation is immediate.
But that’s not all. President Trump truthed that all agencies should immediately cease using Anthropic’s technology. This too has real world effects. Agency heads are appointed by President Trump and will not easily contradict him. Many private companies wish to signal cooperation with this administration. And these effects will very likely not be overturned by an Anthropic legal victory. I’ll cover these at the end.
So what might things look like for the two statutes on May 1st?
Path 1: Courts Block It (~49%)
Five days after the designation, Anthropic filed two simultaneous lawsuits[4]:
The legal arguments are strong. Anthropic’s core claim is that the Pentagon stretched the supply chain risk statute beyond its statutory authority — it was designed for procurement decisions, not as a general-purpose tool to punish a domestic company for its AI ethics policies. There are also First Amendment dimensions: the designation came shortly after Anthropic publicly refused to remove restrictions on mass surveillance and autonomous weapons use. Legal commentary has been blunt — Lawfare called the designation something that “won’t survive first contact with the legal system.”
Base rates
The closest precedent comes from the 2021 wave of “Chinese Military Company” designations under EO 139595 and the 1260H that followed it[5]. The DoD designated dozens of companies with little public explanation. Several sued for Preliminary injunctions (PIs) to hold up the effects until the case was decided in court. All cases where the company sued that I can find details of are below. In the few cases that sought emergency relief, they got it and either won their cases or the Government withdrew.
127 more that didn’t sue.
There are other cases, such as Huawei and Esquel, but those aren’t directly comparable. Huawei was banned by an act of Congress that named it specifically, not by an executive agency decision. Esquel challenged a Commerce Department listing, but the court found Commerce was using that tool exactly as intended. Generally, courts defer much more to Congress than to executive agencies on national security.
The judge
The N.D. Cal case drew Judge Rita F. Lin, a Biden appointee. In Thakur v. Trump, a case involving APA violations and First Amendment retaliation against government contractors, Judge Lin granted a preliminary injunction on a similar legal theory Anthropic is advancing. The 9th Circuit upheld her reasoning on appeal.
That doesn’t guarantee the same outcome. But it seems likely to me. The model puts P(N.D. Cal grants relief) at 83%.
The two-statute problem
Here’s the wrinkle that makes this harder than it looks. The cases are about different statutes, and they live in different courts.
§3252 (DoD-only procurement exclusion) is challenged in N.D. Cal. That’s the case before Judge Lin, and it’s moving fast — hearing scheduled March 24.
§4713 (government-wide exclusion) is challenged in the D.C. Circuit. By statute, challenges to these orders must be filed there. The D.C. Circuit is a separate court with its own timeline.
For the Metaculus question to resolve NO, both statutes need to be blocked. If the N.D. Cal injunction only covers §3252, then §4713 keeps Anthropic locked out of federal agencies. As I’ve mentioned above, this may be the case even if both are overturned, since President Trump has still make his view clear, but the Metaculus resolution criteria seem very reasonable to me.
The key question: could Judge Lin’s N.D. Cal injunction reach broadly enough to cover §4713 too?
Anthropic’s complaint names five counts, including ultra vires Presidential Directive and APA §558 challenges that sweep broadly. The prayer for relief asks the court to block “all Challenged Actions.” But “Challenged Actions” is defined as the Presidential Directive, the Secretarial Order, and actions taken in response — it does not explicitly mention the §4713/FASCSA designation.
I give the probability of the N.D. Cal injunction covering §4713 at 45%. Not impossible — the complaint’s broad framing could pull it in — but the complaint wasn’t drafted to target §4713, presumably deliberately, and Judge Lin might reasonably say that’s the D.C. Circuit’s job.
The D.C. Circuit is the backup. Model gives it a 70% chance of granting relief, but its timeline is slower and less certain. It just might not have judged by May 1st.
Timing
The N. D. Cal hearing is scheduled March 24th. The model expects a ruling within days of that hearing.
I give the D.C. ruling a median of March 30th, 21 days after filing.
The May 1 deadline is 53 days from filing. Xiaomi’s PI came at day 39. Luokung’s at day 62. Over 99% of simulations have at least one court ruling before May 1.
The stay risk
If a court blocks the designation, the government can seek an emergency stay while it appeals. This administration has been aggressive about stays. The model gives an 85% chance the government seeks one.
But getting a stay granted is hard. The combined probability through the 9th Circuit and potential SCOTUS escalation is about 22%[6]. Neither of the Xiaomi/Luokung blocks were successfully stayed. The government would need to show a “reasonable probability of success on appeal,” which is tough when the lower court just found your process arbitrary and capricious.
Path 2: The Government Withdraws (~8%)
The government could pull the designation at any time. In the model, this mostly happens after courts have ruled — either as a face-saving exit after losing (the Xiaomi pattern, where DoD called it a “mutual agreement” ~60 days after the PI) or as a quiet removal (the AMEC pattern, no announcement).
About 8% of simulations end with a government withdrawal. Most of these happen after court setbacks, not before. I think that Anthropic is quite minded to do a deal with the Government, as long as it doesn’t cross their red lines, but I expect most of the probability mass of that after May 1st. The model gives 6.6% unilateral withdrawal, 1.6% withdrawal via deal.
The Polymarket market on “Will Anthropic make a deal with the Pentagon?” prices a deal at about 14% before May 1st (at time of writing). My model’s deal path before May 1st is much smaller.[7] This hasn’t been the central focus of the model, but I’m loath to push it up based on vibes. Anthropic is currently blocked, the courts might rule in the Government’s favour, and there can be stays afterwards and appeals (likely after our deadline). I think unless Anthropic is going to cross their red lines (very unlikely) a deal comes if the Government faces setbacks, not in the next month.
Path 3: The Designation Stands (~43%)
This is the scenario the Metaculus community is roughly pricing in. My simulation agrees it’s substantial.
The biggest YES path isn’t “courts reject Anthropic.” It’s the two-statute structure.
~13% — N.D. Cal rules narrowly in Anthropic’s favour, no DC judgment, §4713 survives. Judge Lin blocks §3252, but her injunction doesn’t reach FASCSA. The D.C. Circuit hasn’t ruled yet, or rules against Anthropic. The government-wide exclusion stays in place. This is the single most likely YES outcome and it’s the one most forecasters seem to be missing.
~12% — Court blocked, but stay granted. A court issues the PI, the government gets an emergency stay reinstating the designation and this runs through May 1st.
~12% — N.D. Cal denies relief outright. Judge Lin rules in favour of the Government, likely on national security grounds. Not the most likely outcome given her track record, but not negligible.
~5% — Both courts deny relief. Everything goes wrong for Anthropic. Both courts find the designation lawful.
~1% — Limbo. A catch all set of weird outcomes. Neither court rules before May 1. Very unlikely given the March 24 hearing.
The Full Monte Carlo
What Metaculus “NO” Actually Means
One thing worth flagging: Metaculus resolving NO does not mean Anthropic gets its government business back. The supply chain risk designation is one of several mechanisms keeping Anthropic out. Even if the courts block it:
The model estimates a 91% chance Anthropic remains effectively blocked even if the designation is lifted.[8] The Metaculus question is a legal question, not a practical one. I can see this changing via a deal (likely after May 1st or the Government losing so decisively that agencies feel comfortable using Anthropic products again. I think these outcomes are very unlikely before the deadline. Note that since there is a 6 month wind-down period, Anthropic might still have Government business but it’s still tailing off.
Metaculus, I was wrong.
The Metaculus community is at 50% YES. My model produces 43%. This seems close enough that I’ll just go with mine.
The original version of this post had a model that returned 14% and contained the following pair of paragraphs:
My bad. Though I still do think this is a bit true. But I’ll defer a bit more in future.
In my defence my forecast was always a blend of my thoughts and Metaculus’ (it never got below 23%). But in this case, I was wrong. I thought that either court could rule on both determinations. Now that seems much less likely.
What I’m Watching
The March 24 hearing. Judge Lin’s line of questioning will signal how she’s reading the case.
The government’s response brief. In Xiaomi, the government’s inability to articulate a factual basis was fatal. I look forward to the DoD’s response.
Whether Judge Lin addresses §4713. If she signals willingness to reach the FASCSA designation, P(CA broad) goes up and YES probability drops.
D.C. Circuit scheduling. Any indication of expedited review there would move the model, making the Metaculus NO resolution more likely (though not changing the overall picture much).
Executive orders. I find President Trump hard to Predict. This model is primarily about the designations, but for a more accurate picture, I am watching his statements on this. There is more the Government can throw at it.
The Bottom Line
My model says 43% chance the designation is still in effect on May 1. And a 91% chance that in some deep sense Anthropic’s business with federal agencies is still tailing off.
Fifty days to find out.
Forecast as of March 12, 2026. Based on a 200,000-simulation Monte Carlo model.
Model details
Model parameters:
Sensitivity analysis (sorted by swing):
Model code available to paid subscribers on request, I intend to write more of these behind a paywall, currently I have a lot of analysis on whether there will be an Iran ceasefire and on silver and oil prices. Thanks Metaculus. This analysis took 2 - 5 hours, and the writing (which involved a lot more analysis) about 8 hours. I would charge somewhere between $1000 and $5000 for analyses like these.
I get $1000, if I beat the community prediction (hard) I get $3500. My hourly rate for this is currently about minimum wage, but I think it's mainly competing with Twitter.
As you can see from the chart, I have updated a lot writing this piece. I will eat my crow at the end.
I think. I have focused my core error correction on the model itself. I can imagine I get one statement like this wrong.
In my view, it is much better to get the overall picture right and the details wrong, than the other way around. I am forecasting the chance these laws are still in effect.
This seems very fast to me, from my rough research. Law in the age of AI.
This is a 134 entity list compiled by Opus 4.6, which I have checked it moderately.
This is the kind of probability I haven't checked that carefully. The sensitivity analysis shows it isn't that important. Am I wrong? Disagree in the comments.
Legal disclaimer. This in not financial advice. Nathan disclaimer. I loath that we live in a society where this is a wise thing to point out.
One might say that this is the lede and that this entire article is confusingly written in regard to that. One also might say that we. are about 7 hours into a 1 hour blog post, so bleh.