Technology Governance

Military AI is governed by contract, not law

In February 2026 the United States designated Anthropic a supply chain risk, an authority reserved until then for foreign adversaries, because the company declined to remove the safety terms from its military contract. No legislature, court, or regulator had been asked to rule on the point first.

P J Laszkowicz

On 27 February 2026, the United States government designated Anthropic, maker of the Claude family of AI models, a supply chain risk. Defense Secretary Pete Hegseth ordered military contractors, suppliers, and partners to sever ties with the company, and President Trump directed every federal agency to stop using its technology, with a phase-out for agencies already embedded in its systems.1

The trigger was not a security breach, a technical failure, or espionage. It was a contractual dispute. Anthropic had refused to remove restrictions in its military contract prohibiting the use of Claude for mass domestic surveillance and fully autonomous weapons targeting.2 Its chief executive, Dario Amodei, said publicly that the company could not "in good conscience" accept the Pentagon's demand for unrestricted access.

Within hours of the designation, OpenAI announced it had secured the same contract for deployment on classified networks.3

Which company made the right decision is not the question here. What matters is what the crisis reveals: a structural contradiction at the centre of American AI strategy, and an institutional vacuum that neither Congress, nor the courts, nor the companies themselves have moved to fill. When a government deregulates an industry to maximise its competitive advantage and simultaneously demands unconditional access to that industry's products, the collision is a matter of when rather than whether. On 27 February it arrived.

A contract, a memo, a deadline

The original terms

In July 2025, Anthropic and the Pentagon signed a two-year agreement worth up to $200 million under which Claude became the first frontier AI model approved for deployment on classified military networks. The deal mattered for its terms as much as its scale. It incorporated Anthropic's acceptable use policy, which prohibited using Claude for mass surveillance of American citizens and for fully autonomous weapons targeting, meaning systems that select and engage targets with no human in the decision loop.4

Vendor-imposed restrictions in a defence contract are unusual without being unprecedented. Contractors have long limited what their products may be used for through export restrictions, end-user certificates, and technology transfer controls. What made Anthropic's restrictions novel is that they were ethical rather than technical or jurisdictional. The company was not saying Claude could not perform these tasks. It was saying Claude should not be used for them.

For seven months the arrangement worked without public friction.

The Hegseth doctrine

On 9 January 2026 Hegseth released two memoranda that redrew the military's relationship with AI vendors. The first, "Artificial Intelligence Strategy for the Department of War", set out AI as a decisive advantage in competition between major powers.5 The second directed that standardised "any lawful use" wording be written into every departmental AI services contract within 180 days.6

The implication was explicit. AI tools bought by the military should carry no vendor restriction beyond what United States law requires. If a use is legal, the government expects the tool to perform it.

These memoranda did not arrive in isolation. They completed more than a year of decisions that had restructured the federal government's relationship with AI development. On 20 January 2025, his first day in office, President Trump rescinded Executive Order 14110, which had established mandatory adversarial testing for high-risk models, cybersecurity protocols, and inter-agency safety collaboration.7 Three days later Executive Order 14179, "Removing Barriers to American Leadership in Artificial Intelligence", replaced it with an action plan built around "sustaining and enhancing the United States' global AI dominance".8 In July 2025 the White House published "Winning the Race: America's AI Action Plan", setting out more than ninety federal actions organised around accelerating innovation, building infrastructure, and leading international AI diplomacy.9 In December 2025 a further order directed the Attorney General to create a task force challenging state AI legislation in court, and conditioned federal grant funding on states avoiding laws the administration considered onerous.10

The memoranda extended that trajectory naturally. Having cleared constraints on AI development, the administration moved to clear constraints on AI deployment.

Twelve days

The public timeline was brief and decisive.

On 15 February, Axios reported that the Pentagon had threatened to terminate the Anthropic contract unless the company removed all acceptable-use restrictions from its military deployment.11

On 24 February, Hegseth delivered a formal ultimatum to Amodei: accept unrestricted use of Claude for all legal purposes by 5.01 pm on Friday 27 February, or lose the contract.12

On 26 February, Amodei answered publicly. Anthropic had negotiated for weeks, he said, and the proposed language "made virtually no progress on preventing Claude's use for mass surveillance of Americans or in fully autonomous weapons", with new wording offered as compromise "paired with legalese that would allow those safeguards to be disregarded at will".13 He affirmed his belief in using AI to defend democracies while holding that a narrow set of uses undermines the democratic values such technology is meant to protect.

Shortly after the deadline passed, Hegseth signed the designation.14 The authority invoked exists to protect the government from foreign-manufactured components carrying backdoors or sabotage mechanisms, and it had been reserved for foreign adversaries such as Huawei and ZTE. This was the first time it was applied to an American company, and it was applied to one that had declined to change its terms of service.

That evening, OpenAI confirmed it had the contract.

The sovereignty paradox

Deregulation as dominance strategy

The policy architecture behind the crisis rests on a coherent strategic logic. The United States leads the world in artificial intelligence because its private sector is innovative, well-capitalised, and lightly regulated. Government's role is to remove barriers so that innovation accelerates. Executive Order 14179 made the logic explicit in its title.

The Stargate Project embodies the approach physically. Announced on 21 January 2025, one day after the inauguration, the $500 billion joint venture between OpenAI, SoftBank, Oracle, and MGX was presented as the infrastructure backbone of American AI supremacy, with SoftBank's Masayoshi Son as chairman and OpenAI holding operational responsibility. The venture projected 100,000 new jobs and an initial deployment of $100 billion.15 By September 2025 OpenAI had announced five further data centre sites, bringing planned capacity to nearly seven gigawatts and more than $400 billion in committed investment.16

This has historical precedent. The internet grew out of the Defense Advanced Research Projects Agency. The satellite positioning constellation was built by the Air Force. The semiconductor industry was catalysed by government purchasing. In each case government created the conditions for private innovation and then took the strategic benefit. AI strategy follows the same pattern.

The pattern carries an assumption: that private companies, empowered and enriched by government policy, will align with government objectives voluntarily. The Anthropic crisis is what happens when that assumption fails.

Access as sovereign prerogative

The other side of the equation is the expectation of access. If government has built the conditions for AI dominance, deregulating the industry, investing hundreds of billions in infrastructure, and clearing legal obstacles at state and federal level, then it expects unrestricted access to the capabilities that dominance produces. From that position, vendor-imposed ethical restrictions on military use are not a reasonable exercise of corporate autonomy. They are an unacceptable constraint on sovereign authority.

The instrument chosen is revealing, and the choice of instrument turns out to matter enormously. Commentary identifies two candidate statutory bases with sharply different consequences. The Federal Acquisition Supply Chain Security Act routes a designation through an interagency council, gives the targeted company thirty days' notice and an opportunity to respond, and provides judicial review in the D.C. Circuit. Section 3252 of Title 10 operates entirely inside the Pentagon, provides no notice, allows no opportunity to respond, and bars judicial review where the government limits disclosure of its determination.17 The designation against Anthropic was effective immediately, issued as a unilateral directive by the Secretary of Defense, with no reported notice or chance to reply.

Legal analysts have noted the incongruity of applying either authority here. Alan Rozenshtein of the University of Minnesota Law School observed that "it's not at all clear that the statute can even apply to an American company where there's no foreign entanglement", noting that these provisions target hostile acts such as sabotage, the malicious introduction of unwanted function, and subversion.18 Lawfare's assessment, published under the title "Pentagon's Anthropic Designation Won't Survive First Contact with the Legal System", argued that the evidentiary requirements for a genuine supply chain risk were plainly not met.

The paradox stated plainly

A government that deregulates to empower private AI companies simultaneously demands those companies surrender the autonomy deregulation was supposed to protect. The company that held its terms lost the contract. The company that accepted the government's terms received it.

This contradiction is not unique to AI, nor to any political party. It is structural, inherent in any arrangement where a government depends on private entities for capabilities it considers essential to national security but does not wish to develop or control directly.

The tension has surfaced before. The Atomic Energy Act of 1946 nationalised nuclear technology precisely to avoid the situation the United States now faces with AI: a critical strategic capability held privately with ambiguous accountability.19 The encryption disputes of the 1990s saw the Clinton administration try to restrict the export of strong cryptography developed by private companies on national security grounds, and lose to a combination of technical reality and constitutional challenge.20 In 2016 Apple refused an FBI demand to build a bypass into the iPhone, setting off a standoff between a company's product security commitments and an agency's investigative needs.21

Each episode turned on the same question: when a private company develops a capability the government considers strategically essential, who sets the terms of its use? The resolutions differed, through nationalisation, legislative compromise, or quiet withdrawal. In none of them was the resolution a unilateral designation of the company as a supply chain risk.

The speed and scale of AI development make the current standoff different in kind. Nuclear technology took decades to mature and stayed largely within state programmes. Encryption was a mathematical discipline with clear technical boundaries. AI is a general-purpose technology developed by a small number of private companies at a pace outstripping any government's capacity to monitor, evaluate, or replicate. The gap between what the government needs and what it can provide for itself has never been wider.

The governance gap

Decisions made bilaterally

Until 27 February, the terms governing American military use of frontier AI were negotiated bilaterally between a Pentagon official and a corporate chief executive. No legislation defined the boundaries. No regulator held jurisdiction. No public deliberation took place. The restrictions Anthropic imposed, and that the government found intolerable, were voluntary corporate policies adopted at the discretion of a private board and enforceable only for as long as that board chose to enforce them.

Democracies do not usually govern strategically consequential technology this way. Nuclear weapons are governed by legislation, by international treaty, and by institutional oversight through the Nuclear Regulatory Commission and the International Atomic Energy Agency. Chemical and biological weapons are governed by their respective conventions. Even conventional arms exports are regulated by the Arms Export Control Act and overseen by the State Department.

For military AI, a technology that may prove more consequential than any of these, the framework consists of executive orders that can be rescinded at will, procurement contracts that can be renegotiated or terminated, and corporate acceptable-use policies that any future chief executive can rewrite.

Georgetown's Center for Security and Emerging Technology has described this as a sovereignty gap in American AI statecraft, finding that critical governance decisions are made through ad hoc executive action and bilateral negotiation, with no institutional durability across administrations.22 The succession from one executive order to its replacement within three days demonstrated exactly that fragility.

The OpenAI question

On 27 February, having secured the contract, Sam Altman announced three red lines in the agreement: no mass domestic surveillance, no autonomous weapons, and no high-stakes fully automated decisions.23

These are substantively identical to the restrictions for which Anthropic was designated a supply chain risk.

The difference is form rather than content. Anthropic's restrictions were embedded in the contract as binding terms, backed by its acceptable use policy and enforceable through the contract's own provisions. OpenAI's were announced in a public statement after signing, and the full contract text was not released.24

The distinction matters. A contractual obligation creates a legal cause of action if breached. A public statement creates a public expectation that can be revised, reinterpreted, or quietly abandoned. The difference between a binding term and a press release is the difference between a lock and a suggestion.

By 3 March, Altman acknowledged the deal "looked opportunistic and sloppy" and said OpenAI was adding to the contract language. Critics noted that the published excerpts were vague and contained potential carve-outs allowing intelligence agencies to conduct surveillance under separate legal authorities.25

OpenAI's own corporate evolution complicates the picture further. On 28 October 2025 the company completed its restructuring as a public benefit corporation, with the original nonprofit retaining a $130 billion stake and Microsoft holding $135 billion.26 Its fiduciary obligations now run, at least in part, to shareholders rather than exclusively to a mission statement about beneficial AI.

What Congress has not done

Across the analytical spectrum one conclusion is nearly unanimous: Congress should legislate.

The Electronic Frontier Foundation argued that privacy protections should not depend on the decisions of a few powerful people, and called for statutory constraints on military AI use and surveillance.27 Rozenshtein made the case more specifically in Lawfare: Congress should define which AI systems the military may buy and on what conditions, set mandatory requirements companies must build into systems sold to government, and impose transparency obligations on buyer and seller alike.28

The International AI Safety Report, published in January 2025 under Yoshua Bengio with contributions from experts across thirty countries, recommended that governments develop independent regulatory bodies for frontier AI rather than relying on industry self-governance or executive discretion.29 The United States has done neither.

The fragility of executive action was demonstrated with unusual clarity in January 2025, when a single presidential signature rescinded the most comprehensive AI safety framework the federal government had produced. Executive Order 14110 was the product of months of inter-agency work, industry consultation, and public comment. It was eliminated in a sentence.30

Without legislation, the rules governing military AI in the United States are whatever the current Secretary of Defense and the chief executive of the winning bidder agree in a given quarter, and they carry no force beyond the tenure of the people who negotiated them.

Concentration and entanglement

The crisis did not occur in a competitive market. It occurred in an oligopoly.

The United States controls approximately 75 per cent of global AI supercomputer capacity.31 Planned American investment in AI infrastructure for 2026, dominated by the Stargate consortium, is projected at $650 billion against Europe's $47 billion.32 The market for frontier general-purpose models is served by a handful of companies, of which Anthropic and OpenAI are two of the most prominent.

Stargate deepens that concentration. A single $500 billion venture binds the government's now-preferred AI company, a major cloud provider, and a foreign capital partner whose chairman also chairs the venture.33 This is not a competitive procurement landscape. It is a structural entanglement between the state and a small number of private actors, carrying dependencies that will be hard to unwind.

The designation showed the practical consequences. Within days, defence contractors including Lockheed Martin began removing Anthropic's technology from their systems.34 Not for any technical deficiency or performance failure, but because of an administrative designation. The government reshaped the commercial AI market through a single executive action, without legislation, judicial review, or public deliberation.

For other AI companies the lesson is plain. Anthropic's safety restrictions were among the best defined and most publicly documented in the industry, consistent with the International AI Safety Report and with the stated policies of every major Western AI company. What set Anthropic apart was not the substance of its restrictions but its willingness to enforce them contractually against its largest customer.

The rational corporate response is to avoid binding restrictions that might trigger similar retaliation. What the precedent establishes is that the cost of enforcing a stated commitment can reach well beyond the loss of a single contract.

Sovereignty beyond America

Three models

The crisis is an American story with global implications. The three major AI powers have each answered the question of who holds sovereignty over artificial intelligence differently.

The United States pursues an industry-led model: private companies develop and control frontier AI, government removes regulatory barriers and acts as a large customer, and governance runs through bilateral negotiation and executive action. The crisis exposed that approach's internal tensions.

The European Union pursues a rights-based regulatory model. The AI Act uses a risk-tiered structure in which certain practices are prohibited outright, including social scoring and real-time biometric surveillance in public spaces, high-risk applications require conformity assessment, and general-purpose model providers face transparency and safety obligations.35 Its provisions phase in across 2025 and 2026 under a newly established AI Office. The framework is prescriptive and institutionally durable, and it will not vanish with a change of government.

China pursues what Beijing calls algorithmic sovereignty, meaning comprehensive state control over AI development, deployment, and the data infrastructure underpinning both. Rather than one comprehensive law, it has issued sector-specific rules covering algorithmic recommendation, generative AI, and synthetic media, each carrying registration, labelling, and security review requirements.36 The People's Liberation Army has reportedly integrated domestically developed models, including DeepSeek, into military systems for non-combat duties, drone swarm decision-making, and battle simulation.37

These are not merely regulatory differences. They reflect incompatible visions of the relationship between state, enterprise, and citizen.

The upstream dependency problem

For the rest of the world the crisis carries a more immediate lesson.

Georgetown's analysis of the sovereignty gap finds that Washington offers allies deployment-layer sovereignty, meaning operational control at the point of use, running on American-built technology. What partner nations increasingly want is independence from upstream dependencies controlled by the United States: the chips that train models, the cloud infrastructure hosting them, and the foundational models themselves.38

The reasons are practical rather than ideological. The American AI Exports Program lacks non-disruption guarantees, exit rights, data residency commitments, and clarity about jurisdictional exposure under the CLOUD Act. European officials have privately described the theoretical ability of Washington to revoke access to AI infrastructure at will.39

The crisis turned that theoretical concern into an observed fact. A single administrative designation removed one of the world's leading AI companies from the entire federal technology stack overnight. Any allied government that had built on Anthropic's technology, or was considering building on any American provider, watched a demonstration of how quickly access can be withdrawn, and on what grounds.

By January 2026 nearly 130 government-backed sovereign AI projects existed across more than fifty countries.40 France had pledged 109 billion euros to sovereign AI infrastructure through 2030, positioning Mistral AI as Europe's alternative to American frontier models.41 India's IndiaAI Mission committed $1.25 billion, including the BharatGen project covering twenty-two languages.42 The sovereign cloud market is projected to grow from $154 billion in 2025 to $823 billion by 2032.43

The European investment gap remains enormous, and independence from upstream American dependencies will be difficult in practice. The strategic motivation has never been clearer, though, and the Anthropic precedent supplies exactly the evidence sovereign AI advocates in allied capitals needed.

What safety means now

Before February 2026 the discourse around AI safety was primarily technical and philosophical. Researchers debated alignment methods, interpretability, evaluation frameworks, and responsible scaling policies. The question was how to make AI systems safe: which techniques, which standards, which benchmarks.

The crisis added a dimension the technical safety community had not fully reckoned with, which is the political economy of safety. Safety restrictions are not free. They carry commercial risk, and that risk is not hypothetical. It can materialise in a single afternoon.

Anthropic's restrictions were not radical. Prohibiting mass domestic surveillance and autonomous weapons targeting is consistent with the International AI Safety Report, with the stated policies of every major Western AI company including OpenAI, and with the principles governments endorsed at the safety summits in Bletchley Park and Seoul.44 Anthropic did not try to prevent the military from using AI. It tried to prevent specific uses it judged incompatible with its safety commitments. The novelty lay in enforcing them contractually against the most powerful customer in the market, and holding that position against an explicit threat.

The chilling effect works at two levels.

At the corporate level, AI companies will hesitate to impose binding safety restrictions on government customers. Voluntary policies may persist in marketing materials and annual reports, where they cost nothing as long as they are never tested against a powerful counterparty's interests. The precedent establishes that contractually enforceable restrictions, the only kind that matter when interests diverge, can carry existential commercial consequences.

At the research level, safety researchers inside AI companies now hold a new piece of information: their work can become a commercial and political liability. The alignment team identifying a risk, the red team documenting a vulnerability, the policy team recommending a restriction, each now knows that its output, if it leads to contractual limits a government customer finds inconvenient, can place the whole company at risk.

Safety work will not stop. The technical problems of alignment, interpretability, and evaluation remain real, and the people working on them are motivated by more than commercial incentive. The institutional conditions have changed, though. The question is no longer only whether alignment can be solved. It is also whether the companies that solve it will be allowed to deploy the solution.

Who decides the limits

The crisis is, as this is written, one week old. Its legal, commercial, and geopolitical consequences will unfold over months and years. Lawfare has argued the designation is unlikely to survive legal challenge. The Electronic Frontier Foundation has called for legislation to fill the vacuum. Defence contractors are rebuilding systems that ran on Anthropic's technology. Allied governments are reassessing their dependence on American AI infrastructure. And the full text of OpenAI's contract, the one that supposedly contains the same restrictions Anthropic lost its contract over, remains unpublished.

The structural question the crisis illuminated precedes this dispute and will outlast it.

In a democratic society, who decides the limits of military AI? Not the limits of what AI can do, which are set by physics and engineering and will keep expanding. The limits of what AI should be permitted to do: which decisions it may take autonomously, which populations it may watch, which parts of targeting and killing may be delegated to a system no human being fully understands.

That question has no answer today—not because it is unanswerable, but because no institution with democratic legitimacy has yet been asked to answer it. Congress has not legislated. The courts have not adjudicated. No regulator has been established. No international framework has been ratified.

Until one is, the answer will keep being set by whoever is in the room when the contract is signed.

Footnotes

  1. ABC News, "Trump orders US government to cut ties with Anthropic AI," 27 February 2026. ↩

  2. TechPolicy.Press, "A Timeline of the Anthropic-Pentagon Dispute," 28 February 2026. ↩

  3. Fortune, "OpenAI sweeps in to snag Pentagon contract after Trump bans Anthropic," 28 February 2026. ↩

  4. Anthropic's acceptable use policy was incorporated by reference into the July 2025 agreement. Terms as described in Amodei's statement of 26 February 2026. ↩

  5. Pete Hegseth, "Artificial Intelligence Strategy for the Department of War," 9 January 2026. ↩

  6. Pete Hegseth, "Transforming the Defense Innovation Ecosystem," 9 January 2026. ↩

  7. Executive Order 14110, signed 30 October 2023, rescinded 20 January 2025. ↩

  8. Executive Order 14179, "Removing Barriers to American Leadership in Artificial Intelligence," 23 January 2025. ↩

  9. "Winning the Race: America's AI Action Plan," The White House, 23 July 2025. ↩

  10. "Ensuring a National Policy Framework for Artificial Intelligence," Executive Order, 11 December 2025. ↩

  11. Ashley Gold, "Exclusive: Pentagon threatens to cut off Anthropic over AI safeguards," Axios, 15 February 2026. ↩

  12. The 5.01 pm deadline is as reported in contemporaneous coverage of the 24 February ultimatum. ↩

  13. Dario Amodei, public statement, 26 February 2026. ↩

  14. An earlier version of this piece gave a precise signing time of 5.14 pm, thirteen minutes after the deadline. That detail could not be confirmed against any source and has been removed; reporting establishes only that the designation followed the deadline the same day. ↩

  15. CBS News, "What is Stargate, Trump's ambitious AI infrastructure venture?" January 2025. ↩

  16. E&E News, "Trump-backed Stargate project unveils plans for 5 AI sites," September 2025. ↩

  17. The procedural contrast between the Federal Acquisition Supply Chain Security Act and 10 U.S.C. § 3252 is set out in Lawfare's analysis. Which authority was used matters a great deal, because only one of them affords notice, a right to respond, and judicial review. ↩

  18. Alan Rozenshtein, University of Minnesota Law School, quoted in Lawfare, 2026. ↩

  19. The Atomic Energy Act of 1946 placed nuclear materials, facilities, and information under the Atomic Energy Commission. ↩

  20. The disputes centred on the Clipper Chip proposal and export controls on strong cryptography, relaxed after legislative pressure, industry lobbying, and Bernstein v. United States (1996). ↩

  21. The FBI's 2016 demand that Apple build a modified operating system to bypass a passcode. Apple refused publicly; the agency accessed the device through a third-party tool, leaving the legal question unresolved. ↩

  22. "The Sovereignty Gap in U.S. AI Statecraft," Center for Security and Emerging Technology, Georgetown University, 2026. The characterisation here paraphrases that analysis rather than quoting it. ↩

  23. Altman's statement was made on 27 February 2026, the same day the contract was announced. An earlier version of this piece dated it to 28 February. ↩

  24. Neither the Anthropic nor the OpenAI contract has been released in full. Analysis of terms relies on published excerpts, company statements, and legal commentary. ↩

  25. TechPolicy.Press, "How to Think About the Anthropic-Pentagon Dispute," March 2026. ↩

  26. TechCrunch, "OpenAI completes its for-profit recapitalization," 28 October 2025. ↩

  27. Electronic Frontier Foundation, 3 March 2026. ↩

  28. Alan Rozenshtein, "Congress—Not the Pentagon or Anthropic—Should Set Military AI Rules," Lawfare, March 2026. ↩

  29. International AI Safety Report, led by Yoshua Bengio, published January 2025. ↩

  30. Executive Order 14110 was rescinded by presidential memorandum on 20 January 2025. ↩

  31. Council on Foreign Relations, "The AI Sovereignty Paradox at Home and Abroad," February 2026. ↩

  32. Ibid. ↩

  33. CBS News, January 2025. ↩

  34. Reuters, "Lockheed, defense contractors removing Anthropic's AI after Trump ban," 4 March 2026. ↩

  35. Regulation (EU) 2024/1689. Prohibitions took effect 2 February 2025; general-purpose model obligations 2 August 2025; remaining provisions from 2 August 2026. ↩

  36. China's approach includes the Algorithm Recommendation Regulation (2022), the Deep Synthesis Regulation (2023), and the Generative AI Measures (2023). ↩

  37. The Defense Post, "China Accelerates Military AI Drive With DeepSeek Model," 29 October 2025. ↩

  38. Center for Security and Emerging Technology, 2026. ↩

  39. Ibid. The CLOUD Act allows US law enforcement to compel disclosure of data held by US-headquartered companies wherever it is physically stored. ↩

  40. Lawfare, "Sovereign AI in a Hybrid World: National Strategies and Policy Responses," 2026. ↩

  41. McKinsey, "Sovereign AI: Building Ecosystems for Strategic Resilience and Impact," 2025. ↩

  42. International Institute for Strategic Studies, "Sovereign AI: Pathways to Strategic Autonomy," August 2025. ↩

  43. Lawfare, 2026, citing industry projections. ↩

  44. The International AI Safety Report recommended against deploying AI in fully autonomous weapons without meaningful human control, and called for restrictions on AI-enabled mass surveillance. ↩