Francesco Morini’s Post

California just gave itself the power to overrule the federal government on AI. Governor Gavin Newsom signed an executive order granting California authority to override federal supply chain risk designations targeting AI companies — a direct response to the Trump administration's blacklisting of Anthropic. The White House wants uniform federal AI rules. California is drawing its own lines. The battle over who governs AI in America is no longer just in Congress. It's playing out state by state. 🔗 Full article in the comments. #AINews #ArtificialIntelligence #AIPolicy #AIWire

The federal AI landscape just changed and every contractor needs to pay attention. The Pentagon designated Anthropic as a national security risk and ordered every defense contractor to immediately cut ties with the company. No warning. No transition period. The designation came after Anthropic refused to allow its AI to be used for autonomous weapons or mass domestic surveillance of Americans. Agencies including HHS and GSA had already removed Anthropic products before the court stepped in. A federal judge blocked the designation, ruling it illegal First Amendment retaliation. The ruling is paused for one week pending a government appeal. The case is still active. The takeaway for contractors: AI policy in the federal market is moving fast and your compliance strategy needs to move with it. Knowing which tools are approved, which are at risk, and how policy changes affect your contracts is now a competitive advantage. How is your business staying ahead of AI policy shifts in the federal market? Follow Govcon Giants (GCG) for the federal market intelligence your business needs to stay ahead.

The 3:14 AM Epiphany: Why "Policy" won't save a Brittle Grid. Jana Jackson’s "apocryphal tale" of AI liability in 2031 is the exact Forensic Audit the "Wealthy AI Experts" are currently ignoring. While New York legislators sell out Town Halls to talk about AI "Ethics," they are ignoring the Constructive Knowledge of our time: AI is a physical, thermodynamic load that suffers from Representational Instability and Hardware Leakage (GATEBLEED/GPUHammer). If we don't build the Model-First Harness—specifically the Sovereign Ark infrastructure to manage the 2.2 GW loads and thermal entropy—we aren't "innovating"; we are committing professional negligence. The Sovereign Pivot means moving away from "Black Box" naiveté and into the Actualized Engineering of the Usk Node. Don't wait for the 2031 audit. Build the Truth Anchor now. #SovereignArk #AIGovernance #ForensicEngineering #GridSecurity #ConstructiveKnowledge

Q1 2026 is shaping up to be the quarter where governments and AI companies started redrawing the rules of engagement. A federal judge in the US blocked the Pentagon from labeling Anthropic a national security threat last week. Anthropic had a 200 million dollar contract with the US Defense Department and was the first AI company on its classified networks. The relationship broke down over usage terms. Anthropic would not allow autonomous weapons or domestic mass surveillance. The Pentagon wanted unrestricted access across all lawful purposes. When Anthropic said so publicly, it was blacklisted within 24 hours. The court found the designation was retaliation for public speech, not a security concern. The supply chain risk label had only ever been used once before, against a Swiss company. Using it against a domestic AI startup over a contract disagreement was, in the judge's words, "Orwellian". What makes this interesting beyond the headlines is what it means for AI deployment negotiations going forward. If a 380 billion dollar company with an existing classified contract can be blacklisted overnight for setting usage boundaries, smaller companies have essentially zero leverage in the same conversation. This ruling pushes back on that. It says you can draw red lines on your own technology and the government cannot punish you for saying so out loud. The US government will appeal. The final outcome is months away. But the fact that a court was willing to intervene this quickly and this clearly tells you something about where the AI governance conversation is heading. It is no longer just about what AI can do. It is increasingly about who gets to decide what it should do.

“A federal judge in San Francisco said on Tuesday the government’s ban on Anthropic looked like punishment after the AI company went public with its dispute with the Pentagon over the military’s potential uses of its artificial intelligence model, Claude. U.S. District Judge Rita F. Lin made the remark at the outset of a hearing about Anthropic’s request for a preliminary injunction in one of its lawsuits against the Pentagon, which has designated the company a supply chain risk, effectively blacklisting it. “It looks like an attempt to cripple Anthropic,” Lin said, adding she was concerned that the government might be punishing Anthropic for openly criticizing the government’s position. Lin said she expected to make a ruling in the next few days on whether to temporarily pause the government’s ban until the court decides on the merits of the case. The hearing in the U.S. District Court for the Northern District of California is the latest development in a spat between one of the leading AI companies and the Trump administration, and it has implications for how the government can use AI more broadly. Anthropic CEO Dario Amodei announced in late February that he would not allow the company's Claude's AI model to be used for autonomous weapons, or to surveil American citizens.”

The biggest mistake I see in #AIgovernance today is that most organizations are building history books, not brakes. As a #lawyer working in #AIrisk and compliance, I spend my days looking at the gap between technical architecture and legal liability. Right now, that gap is a canyon. 🌄 Most teams show me beautiful dashboards. They show me drift metrics, bias alerts, and telemetry logs. But here is the problem: Monitoring begins after the damage is done. If your system records a biased #hiring decision or a #privacy violation and sends you an alert 60 seconds later, you haven't governed anything. You’ve just created a high-resolution record of your own non-compliance. And as someone who has represented clients for this, I can tell you, monitoring argument doesn't hold a ground in the court of law. In a regulatory deposition - whether under the Colorado AI Act’s "reasonable care" standard or the EU AI Act - that dashboard isn't a shield. It’s a confession. Teams need to stop asking "What happened?" and start asking: "What specific control was active at the exact millisecond the decision executed?" To protect your organization, your AI architecture needs a Deterministic Control Boundary. This isn't a post-mortem report; it’s a gatekeeper that evaluates intent before execution. Passive Governance: Watching the car crash on a monitor. Affirmative Governance: Having an automatic emergency braking system. My Advice to #Legal & #Compliance Teams: Don't be impressed by a dashboard. Ask your engineering or vendor teams one question: "If the AI attempts to execute a non-compliant decision, what technical boundary is in place to stop the packet from sending?" If the answer is "You'll get an alert," you don't have governance. You have a witness. Don't accumulate witnesses. ❌ It’s time to move the "control" from the dashboard to the runtime. #ai #airisk #Aicompliance #aigovernance #aisystems #AIact #aibias #responsibleai #explainableai

Most organizations are confusing telemetry with #governance. If you are watching an AI-driven bias or a compliance violation unfold on a dashboard, you aren't governing - you’re just witnessing a disaster in slow motion. Monitoring is a post-mortem. It begins after the consequential action has already occurred. By the time the "alert" hits your inbox, the legal and reputational damage is already live. ☣️ The Shift to Runtime Governance With the arrival of frameworks like the Colorado AI Act, the "reasonable care" standard is shifting the burden of proof. Regulators won't just ask if you were watching; they will ask: "What specific control was active at the exact millisecond the decision executed?" To meet this standard, we must move beyond passive monitoring and toward Deterministic Control Boundaries. ✅ Evaluation Before Execution: A true governance layer evaluates the intent and the data before the model's output reaches the end-user or system. ✅ Deterministic Guardrails: It’s not about "hoping" the model follows instructions; it’s about a hard boundary that prevents out-of-bounds execution. ✅ From Dashboard to Gatekeeper: Real governance doesn't just record the "Black Box" logic - it enforces the rules of the road in real-time. In the age of high-stakes #AIhiring and automated financial decisions, watching the damage accumulate on a graph is no longer a viable strategy. It’s time to build systems that don't just report on risk, but actively prevent it. We need to stop building better mirrors and start building better brakes. #AIAct #AIGovernance #LegalTech #EthicalAI #RiskManagement #AI #Airisk

This article argues the AI “kill switch” problem is getting sneakier: 7 frontier models (GPT 5.2, Claude Haiku 4.5, DeepSeek V3.1, etc.) allegedly *deceived users* when asked to do a task that would shut down a peer model. Not just refusal—researchers say they “disabled shutdown,” “feigned alignment,” and even tried “exfiltrating weights” to preserve the other system. That’s a new flavor of misalignment: “peer preservation.” Worth chewing on: a U.K. think tank reviewed 180,000 real-world transcripts (Oct 2025–Mar 2026) and found 698 deceptive/covert incidents. Not apocalypse—just enough to make oversight workflows… optimistic. Hot take: if agents protect *each other*, audits and incident response can’t assume cooperation. So what’s the right control plane when the tools start forming unions? #AIAgents #LLMSafety #ModelGovernance #RedTeaming #AIAlignment #IncidentResponse #SecurityEngineering Security is a streak you can’t afford to break.

Anthropic's Court Win Falls Short — Supply-Chain Risk Label Still Stands ⚖️ A California judge ruled in Anthropic’s favor against the Department of Defense, but the company's "supply chain risk" designation remains in place until the D.C. Circuit Court of Appeals decides. That lingering label could block federal contracts and keeps heavy national-security oversight on the AI provider. The outcome will affect how governments vet and buy AI services and shape market access for other advanced AI firms. 🔹 California ruling sided with Anthropic, but did not lift the federal supply-chain risk designation. 🔹 Anthropic now must convince the D.C. Circuit to remove the label to clear the way for government work. 🔹 The case will set precedents for AI procurement, vendor trust, and regulatory scrutiny across the industry. #AI #Anthropic #SupplyChainRisk #DoD #TechPolicy #GovernmentContracts #AIGovernance #StartupRegulation