The Vendor Reckoning

On Feb 28, the Trump administration blacklisted Anthropic after the company refused to remove ethical safeguards from military AI systems — specifically constraints on autonomous weapons and mass surveillance. All federal contracts were terminated. Federal agencies were ordered to drop Claude within six months. The same day, OpenAI secured Pentagon access with guardrails prohibiting autonomous weapons direction and unconstrained surveillance — constraints similar to Anthropic's, negotiated rather than imposed.

The immediate effect is a two-tier federal AI market: vendors willing to negotiate safeguards within government parameters, and vendors maintaining constraints the government finds unacceptable.

But the week's data complicates the narrative. Anthropic raised $30B at a $380B valuation (up from $183B five months prior at Series F) — private capital treating the blacklist as irrelevant, or possibly as a trust signal. Claude ranked #1 on the US App Store, surpassing ChatGPT and Gemini. And reports that CENTCOM continued deploying Claude after the ban — though the sourcing is thin and should be treated with caution — suggest that operational embedding resists political override on short timescales.

The counter-case is real: government procurement bifurcation has historical precedent. During the Cold War, US restrictions on crypto exports barred vendors from federal channels, yet those companies survived and dominated civilian markets. Anthropic's refusal may simply accelerate private-sector adoption while OpenAI captures federal dollars — a market segmentation, not a market failure.

---

Block cut over 4,000 jobs — roughly 40% of its ~10,000-person workforce — with CEO Jack Dorsey explicitly citing AI for efficiency gains. Stock rose over 20%. Same week, IBM fell 13% after Claude Code raised questions about COBOL modernisation revenue — IBM's mainframe services business has long depended on COBOL expertise as a moat. Thomson Reuters rebounded 11% (recovering from a 26% year-to-date decline) after earnings beat expectations and Anthropic confirmed Claude-powered agents in Thomson Reuters' workflow — a combination of fundamentals and AI narrative.

Yet Goldman Sachs reported AI contributed roughly zero to US GDP in 2025. That finding comes with a caveat: Goldman's methodology measured AI-specific investment contribution to aggregate GDP, not firm-level productivity — a metric that may not capture micro-level efficiency gains that haven't yet scaled to macro effect.

The counter-case: stock rises on layoff announcements are not new — it's a familiar pattern when markets interpret cuts as margin discipline. Dorsey's AI attribution may signal strategic confidence rather than a structural threshold. And a 40% headcount cut typically requires rehiring in AI operations, data engineering, and model deployment — net workforce reduction may be materially smaller than the headline.

---

Australia's eSafety Commissioner set March 9 — seven days from publication — as a hard deadline for AI age verification compliance. Non-compliant apps face app store removal. The mechanism is specific: the regulator is threatening Apple and Google directly, shifting enforcement from app developers to platform gatekeepers. If app stores comply, the removal cascade is automatic. If they don't, Australia's enforcement credibility is tested immediately.

This is what makes March 9 different from prior regulatory announcements. It's not a framework. It's not a consultation period. It's a date, with a consequence, aimed at two companies (Apple and Google) that have operational capacity to enforce it overnight.

The counter-case: Australia's social media minimum age law (passed December 2024) set a December 2025 enforcement date — and the eSafety Commissioner's initial response was regulatory guidance rather than immediate penalties. The agency's annual base budget (~A$42.5M, quadrupled in 2023 from A$10.3M) still limits its capacity for simultaneous enforcement across multiple fronts. Google negotiated commercial deals under Australia's News Media Bargaining Code specifically to avoid the arbitration mechanism — a pattern of platform negotiation rather than outright compliance. The March 9 deadline may produce a handful of high-profile removals rather than the systematic enforcement the announcement implies.

---

The Australia deadline isn't the only regulatory threshold crossing this week. Inside the US, a quieter but structurally more significant conflict is taking shape.

States across the political spectrum are limiting AI in health insurance underwriting — red states concerned about algorithmic pricing, blue states concerned about discrimination. Trump responded with an executive order seeking to preempt state AI rules entirely. Utah advanced its AI Transparency Act, requiring frontier developers to publish safety and child protection plans.

Pulse24's analysis: the federal-state tension is structurally different from Australia's top-down deadline. Australia has one regulator, one deadline, one mechanism. The US has 50 states legislating independently, a federal executive attempting blanket preemption, and no judicial resolution in sight. For companies deploying AI across state lines, the compliance question isn't "which rule applies" — it's "which rules contradict each other, and which jurisdiction will enforce first."

This matters for AI vendors specifically because the insurance AI restrictions target algorithmic decision-making — the same capability that underpins underwriting, lending, and hiring tools. If state-level AI restrictions survive federal preemption, the precedent extends well beyond health insurance.

---