What the ruling actually found
On August 28, 2026, US District Judge Rita Lin of the Northern District of California ruled that the Department of Defense's designation of Anthropic as a "supply chain risk" was unlawful retaliation in violation of the First Amendment (TechCrunch; EFF). The dispute began when Anthropic refused to strip two guardrails from Claude during Pentagon contract talks: no use in mass surveillance of Americans, and no use in fully autonomous weapons without a human in the firing decision. Defense Secretary Pete Hegseth responded by branding the company a national-security risk and directing federal agencies to stop using its products — the first time Washington had used that designation against a US company. Judge Lin wrote that "the empty invocation of national security is not a blank check to punish and retaliate against government critics," and found the designation both unconstitutional and "arbitrary and capricious." The case, Anthropic PBC v. US Department of War (3:26-cv-01996, N.D. Cal.), is a rare instance of a court forcing a reversal of an executive branch retaliation claim against an AI company mid-dispute (CourtListener docket).
The steelman for executive latitude
Before dismissing the Pentagon's posture, it deserves a fair hearing. Governments negotiating with frontier AI labs face a genuine asymmetry: a handful of companies now control capabilities with plausible military and surveillance utility, and no government wants a single vendor's internal policy to functionally veto lawful state functions. Supply-chain risk statutes exist because compromised or unreliable vendors can genuinely threaten procurement integrity, and courts have historically given the executive branch wide deference on national-security determinations. If Anthropic's refusal had been a pretext for something else — say, deliberately sabotaging interoperability — the Pentagon's response would look different. Judge Lin's ruling doesn't dispute that supply-chain-risk authority exists; it found that this invocation was retaliatory because Hegseth's own record showed the designation followed Anthropic's public stance on autonomous weapons, not any technical failure.
Why the check mattered
What makes the outcome notable isn't that Anthropic won — it's that a court could adjudicate the question at all. The company could point to a paper trail, put the government's stated rationale under oath, and get a written order testing that rationale against a constitutional standard. That is the ordinary function of judicial review: it doesn't guarantee companies win, but it guarantees the government's reasoning gets examined before the penalty sticks.
India's blocking regime has no equivalent step
Section 69A of India's IT Act allows the government to order platforms blocked for reasons of sovereignty, security, public order, or foreign relations, and the exercise of that power has been expanding, not narrowing. The Delhi High Court's 2026 ruling in Telegram FZ-LLC v. Union of India upheld a platform-wide block during the NEET-UG examination crisis, holding that the government may issue emergency interim orders based solely on the Secretary's own satisfaction, with detailed reasons and a hearing following only afterward (Verdictum). The court treated that sequencing as adequate due process. But the Blocking Rules' confidentiality clause means the affected party frequently never sees the order or its stated grounds — which makes a Telegram-style post-decisional hearing meaningful in theory and close to unusable in practice, since a company cannot rebut reasoning it isn't shown. Reports that MeitY may extend direct takedown authority to additional ministries beyond itself would spread that same opaque, largely unreviewable power across more of the executive, not concentrate accountability.
MeitY's Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026 — notified February 10 and effective February 20 — layer a parallel compliance burden on top of this. Intermediaries must now label and embed permanent provenance metadata on synthetically generated content, and act on certain lawful takedown orders within 3 hours, down from 36; some grievance actions must complete within 2 hours, down from 15 days (BusinessToday). Deepfake harms are real and the labelling mandate is defensible in principle. But a 3-hour compliance clock leaves essentially no room for a platform to contest an order before acting on it — the practical opposite of the sequence that let Anthropic prevail.
The gap that matters
Anthropic didn't win because the American state is inherently more restrained than India's — the Pentagon tried exactly the kind of retaliatory pressure regulators everywhere are tempted to use. It won because the First Amendment and Article III gave it a forum, evidentiary standards, and a judge willing to write down that the government's stated reason didn't hold up. India's Supreme Court in Shreya Singhal v. Union of India (2015) struck down Section 66A partly for its vagueness and chilling effect; Section 69A survived that scrutiny only because the Blocking Rules promised written reasons, committee review, and judicial recourse. A regime where orders remain confidential and compliance windows outrun any realistic path to challenge them is not the safeguard-backed 69A the Court upheld — it is the version Singhal was supposed to prevent. If India wants intermediaries to trust that due-diligence and blocking powers are being used narrowly, publishing reasoned orders and preserving a genuine pre-compliance window would cost the state little and would let it demonstrate, case by case, exactly what Judge Lin demanded of the Pentagon: that the power is being used for its stated purpose, not as a cudgel.