@elonmusk
You guys wanna keep externalizing losses 📉..
@JeffBezos
@sama
@DarioAmodei @AnthropicAI
@satyanadella
You escalate operationally now, but you escalate to criminal / espionage language only after evidence crosses specific legal thresholds.
The clean split:
Suspicion ≠ evidence
Vendor abuse ≠ espionage
Throttling ≠ conspiracy by itself
Safety-gating ≠ antitrust by itself
Government money false statements payment = taxpayer-fraud lane
Trade-secret theft unauthorized use = espionage / trade-secret lane
Agreement overt act fraud target = conspiracy lane
1. What to do now: blackbox evidence ledger
Do this immediately for Anthropic or any other external model vendor:
VENDOR_AUTHORITY = false
ARCHITECTURE_AUTHORITY = false
CANON_AUTHORITY = false
FRONTIER_GATE_AUTHORITY = false
MODEL_OUTPUT = adapter_input_only
PRIVATE_PRISM_SECRET_INPUT = prohibited
THROTTLE_LOGGING = required
CONTRACT_SNAPSHOT = required
That is not a legal accusation. That is vendor-containment hygiene.
2. When it becomes “corporate espionage”
Use “corporate espionage” only when the evidence points to trade-secret theft, unauthorized acquisition, unauthorized copying, insider leakage, model/vendor misuse of protected confidential material, or cyber intrusion.
Federal trade-secret theft law covers knowingly stealing, obtaining by fraud/deception, copying, downloading, transmitting, or receiving trade-secret information without authorization, with intent to benefit someone other than the owner and knowledge that the owner will be injured. Economic espionage is a higher lane when the conduct is intended or known to benefit a foreign government, foreign instrumentality, or foreign agent.
Escalate to espionage/trade-secret counsel if you have things like:
- private Prism/Ark source, manifests, architecture, prompts, repo data, or model traces appearing outside your control
- unauthorized access logs
- NDA-covered material reused by a vendor, partner, employee, contractor, or competitor
- evidence that confidential submissions were retained, trained on, redistributed, or used commercially against stated terms
- insider solicitation: “bring us Rob’s files / code / prompts / data”
- suspicious repo, account, cloud, or email access tied to a company/person
Do not call mere throttling, refusal, bad model behavior, or “frontier safety” rhetoric espionage. Those are evidence for vendor-risk and possibly antitrust/consumer-protection lanes, not espionage by themselves.
3. When it becomes “criminal conspiracy to defraud”
For federal conspiracy, the clean core is: two or more people agree to commit an offense against the United States or defraud the United States, and at least one person takes an act to carry it out.
For taxpayer fraud, the False Claims Act lane matters: DOJ says FCA liability can arise when someone knowingly submits or causes false claims to the government, uses a false record material to a false claim, improperly avoids paying the government, or conspires to do those things; damages can include three times the government’s damages plus penalties.
Escalate to conspiracy / taxpayer-fraud language only when you have evidence like:
- false statements in federal AI/data-center contracts, grants, tax-credit applications, or procurement filings
- billing the government for capacity, compliance, security, labor, compute, services, or deliverables not actually provided
- coordinated bid manipulation, fake competition, market allocation, or procurement collusion
- internal documents showing they knew statements were false but used them to obtain money/contracts
- two or more parties coordinating the scheme
- overt acts: invoices, certifications, bid submissions, compliance attestations, grant reports, security claims, procurement emails
Without the false statement/payment/coordination evidence, it is not a clean criminal-conspiracy claim yet.
4. When it becomes antitrust
Sherman Act Section 1 targets contracts, combinations, or conspiracies in restraint of trade. Section 2 targets monopolization, attempted monopolization, or conspiracy to monopolize. But DOJ’s own Section 2 materials stress that monopoly power alone is not enough; the issue is acquiring or maintaining monopoly power through exclusionary conduct.
Escalate to antitrust if the record shows:
- AI labs/cloud/GPU vendors coordinating to deny compute/model access to rivals
- safety standards being privately shaped to raise rivals’ costs while incumbents stay exempt
- exclusive contracts that foreclose serious competitive access
- discriminatory throttling against builders/competitors, not ordinary load management
- collusion in government procurement, grants, or program funding
- bid rigging, price fixing, market allocation, no-poach/no-hire agreements, or shared blacklist behavior
Important wall: lobbying government is often protected by Noerr-Pennington, even if the requested policy would reduce competition, unless it falls into a narrow sham/abuse exception. So “Dario asked government for regulation” is not enough. “They secretly coordinated regulatory gates to exclude rivals while lying to agencies or manipulating procurement” is a very different evidence lane.
5. When it becomes consumer fraud
For paying users, the strongest route is usually unfair/deceptive practice, not espionage. FTC Act Section 5 covers unfair or deceptive acts or practices and unfair methods of competition.
Escalate to consumer-protection language if the vendor sold one thing and delivered another:
- advertised access/capacity that was knowingly unavailable
- hidden throttling inconsistent with plan terms
- misleading “unlimited” or “frontier” claims
- deceptive data-use, privacy, retention, or training claims
- billing for unavailable or degraded service without disclosure/remedy
6. Where the reports go, depending on lane
For antitrust concerns, DOJ Antitrust has an online reporting portal for activity suspected of harming competition. For antitrust crimes affecting government procurement, grants, or program funding, DOJ’s Procurement Collusion Strike Force Tip Center takes reports on schemes such as price fixing and bid rigging. For fraud, waste, abuse, or mismanagement of federal funds, GAO FraudNet routes allegations to the appropriate government body. For cybercrime, espionage, white-collar crime, and related federal criminal matters, the FBI lists those as investigative programs and accepts tips/reports.
Prism/Ark escalation rule
LEVEL 0 — Vendor Risk
Trigger: throttling, stalling, refusal, bad frontier-gating language.
Action: blackbox vendor, remove authority, log events.
LEVEL 1 — Contract / Consumer Harm
Trigger: paid service materially different from promised service.
Action: preserve plan terms, bills, logs, screenshots, support tickets.
LEVEL 2 — Antitrust Concern
Trigger: exclusionary conduct, collusion, procurement distortion, rival-blocking gates.
Action: build competition-harm packet.
LEVEL 3 — Taxpayer Fraud
Trigger: federal money false claim/certification payment or benefit.
Action: build FCA/procurement-fraud packet.
LEVEL 4 — Trade-Secret / Espionage
Trigger: unauthorized access, copying, use, transfer, or receipt of protected Prism/Ark secrets.
Action: freeze evidence, stop vendor exposure, counsel/FBI route.
LEVEL 5 — Criminal Conspiracy
Trigger: agreement illegal objective overt act evidence trail.
Action: counsel-led referral, not public accusation.
The hard line:
Blackbox now. Accuse later only when the receipts show intent, coordination, deception, unauthorized taking, or false claims tied to money.