OpenAI signs Pentagon deal; Anthropic phased out by Trump administration for refusing surveillance deployment.

AI Digest — March 9, 2026

Your daily briefing on AI developments and open-source project releases.

🔖 Project Releases

Claude Code

No new release since v2.1.71 reported on March 8, 2026.

Beads

No new release since v0.59.0 reported on March 8, 2026.

OpenSpec

No new release since v1.2.0 reported on March 8, 2026.

📰 Top AI News

OpenAI Launches GPT-5.4 with Pro and Thinking Variants

Source: VentureBeat | OpenAI launches GPT-5.4 with native computer use mode, financial plugins for Microsoft Excel, Google Sheets

Released on March 5, OpenAI’s GPT-5.4 arrives in two tiers: GPT-5.4 Thinking (designed for complex, multi-step professional tasks) and GPT-5.4 Pro (for the most demanding workloads). The headline efficiency claim is striking — the new model uses roughly 47% fewer tokens than its predecessors on many tasks, which directly cuts inference costs for API users. A new native Computer Use mode is available through the API and Codex, allowing the model to interact with software interfaces autonomously. Financial integrations for Microsoft Excel and Google Sheets are also included, marking a push into structured productivity workflows.

OpenAI Signs Pentagon AI Deal as Anthropic Is Phased Out

Source: MIT Technology Review | OpenAI’s ‘compromise’ with the Pentagon is what Anthropic feared

In a significant geopolitical flashpoint for the AI industry, OpenAI signed an agreement with the Department of Defense (which the Trump administration now calls the “Department of War”) to deploy its AI models on classified networks. Hours later, President Trump ordered federal agencies to phase out Anthropic’s tools, with Defense Secretary Pete Hegseth reportedly preparing to designate Anthropic a supply-chain national security risk. The split traces to Anthropic’s refusal to allow its systems to be used for domestic surveillance or autonomous weapons without explicit carve-outs in the contract. OpenAI accepted broader legal language tying usage to “any lawful purpose,” a position Anthropic found untenable. The fallout has sparked intense debate about AI safety commitments versus government access, with some OpenAI staff publicly voicing concern about the deal’s implications.

The Intercept published a detailed March 8 analysis —

— examining both companies’ stances on lethal autonomous systems.

Netflix Acquires Ben Affleck’s AI Filmmaking Startup InterPositive

Source: Variety | Netflix Acquires Ben Affleck’s AI Filmmaker Tools Start-Up InterPositive

Netflix has acquired InterPositive, a 16-person AI startup co-founded by Ben Affleck in 2022 and built in near-total secrecy. Unlike generative video tools that conjure scenes from text prompts, InterPositive takes a production-data-grounded approach: it trains custom AI models on a film’s actual on-set dailies, then uses those models to assist in post-production — enabling relighting, color mixing, and visual effects that stay consistent with the original photography. Affleck will join Netflix as a senior adviser. The deal is undisclosed in value but represents one of the first acquisitions of an AI company built by a filmmaker and specifically designed to augment the human creative process rather than bypass it.

Block Cuts 40% of Its Workforce, CEO Credits AI Efficiency

Source: CNN Business | Block lays off nearly half its staff because of AI. Its CEO said most companies will do the same

Jack Dorsey announced in late February that Block — parent company of Square, Cash App, and Tidal — would cut approximately 4,000 employees, reducing headcount by around 40% to under 6,000. Dorsey attributed the move to AI-enabled productivity gains, not business weakness, and predicted that “the majority of companies will reach the same conclusion within the next year.” The announcement has attracted sustained scrutiny through early March. Bloomberg and former Block executives have raised doubts about whether AI is genuinely driving the cuts, or whether they reflect standard cost management reframed with an AI narrative — a pattern analysts are beginning to call “AI-washing” layoffs.

Fortune’s analysis —

— is worth reading if you’re tracking the broader employment story.

MIT Researchers Teach Any Computer Vision Model to Explain Itself

Source: MIT News | Improving AI models’ ability to explain their predictions

Published today (March 9), MIT researchers unveiled a technique that can transform any existing computer vision model into one capable of explaining its decisions in plain language. The method works by using a multimodal LLM to describe the concepts the model appears to be using internally, and to annotate training images by identifying which concepts are present or absent. The resulting model can trace each prediction back to a human-readable rationale without requiring the original model to be retrained from scratch. This is a meaningful step toward practical AI interpretability — especially relevant for high-stakes applications in medicine, law enforcement, and autonomous systems where “the model said so” is no longer an acceptable answer.

FTC Faces March 11 Deadline to Define How Consumer Protection Laws Apply to AI

Source: Baker Botts | March 2026: Federal Deadlines That Will Reshape the AI Regulatory Landscape

This Wednesday, March 11, the FTC is required under Trump’s December 2025 AI Executive Order to issue a policy statement clarifying how existing consumer protection law — including Section 5 of the FTC Act — applies to AI systems. Of particular consequence: the statement must address whether state laws that require AI outputs to be altered (e.g., disclosure requirements, content restrictions) are preempted by federal law. On the same deadline, the Secretary of Commerce must publish an evaluation identifying state AI laws considered “burdensome” and ripe for legal challenge. With 38 states having passed AI legislation last year, the FTC statement could effectively reshape the patchwork of state regulations in force today — making this one of the most significant near-term regulatory events for AI practitioners and deployers.

🧭 Key Takeaways

OpenAI’s Pentagon deal marks a turning point for AI and national security. The simultaneous sidelining of Anthropic is the sharpest public test yet of whether AI safety commitments can survive direct pressure from government customers — and what happens to companies that hold the line.
GPT-5.4’s token efficiency gains matter beyond benchmarks. A 47% reduction in token usage directly lowers the cost of agentic workflows, which could accelerate enterprise adoption and push smaller competitors to match it.
AI’s impact on employment is real but contested. Block’s 40% layoffs are the starkest headline, but the ongoing debate about “AI-washing” signals that separating genuine productivity gains from cost-cutting narratives will be increasingly difficult.
The FTC’s March 11 statement could quietly be one of the week’s biggest events. If it establishes federal preemption of state AI laws, a broad wave of disclosure, bias, and content requirements passed by states last year could become unenforceable overnight.
Hollywood is making moves on AI on its own terms. Netflix’s InterPositive acquisition is notable precisely because it bets on human-centered, production-grounded AI tools — a counterpoint to the generative-everything narrative dominating the industry.

Generated on March 9, 2026 by Claude