AI-Proof - Weekly AI Pulse
A concise summary of the week’s most important AI developments
Executive Summary
This was the week AI became a political weapon.
The Trump administration blacklisted Anthropic as a “supply chain risk to national security” - a designation previously reserved for hostile foreign states - after the company refused to remove guardrails on mass surveillance and autonomous weapons from its Pentagon contract. OpenAI swooped in hours later with its own classified-network deal, triggering the largest consumer backlash in AI history: ChatGPT uninstalls surged 295%, Claude hit number one on the App Store, and hundreds of tech workers across Silicon Valley signed open letters opposing unchecked military AI use.
Meanwhile, OpenAI closed a $110 billion funding round - the largest private raise in history, and Nvidia committed $4 billion to photonics companies to secure the optical interconnects that will power next-generation AI data centres. Big Tech is now forecast to spend $650 billion on AI infrastructure in 2026 alone.
On the model front, OpenAI shipped GPT-5.3 Instant with a 27% hallucination reduction, Google launched Gemini 3.1 Flash-Lite at a tenth the cost of Pro, and DeepSeek V4, a trillion-parameter open-weight model from China, is expected imminently. For businesses, the week crystallised three uncomfortable truths: AI is now inseparable from geopolitics, capital concentration is accelerating, and the model release cadence has become a permanent sprint.
Why this matters to businesses
Your AI vendor choices now carry political risk - The Anthropic blacklisting is a wake-up call. Your choice of AI provider is no longer just a technology decision - it is a supply chain risk decision with geopolitical exposure. If your business relies on a single model provider and that provider falls out of favour with a government you operate under, your access could be disrupted overnight. Diversify your model dependencies now.
The consumer backlash is a leading indicator for enterprise The 295% surge in ChatGPT uninstalls was consumer-driven, but enterprise buyers are watching closely. Companies seen as ethically compromised in AI face reputational risk with customers, employees, and partners. Vendor values on safety and governance are now material considerations, not marketing fluff.
The cost of inference is collapsing - plan accordingly Google’s Flash-Lite at $0.25 per million input tokens is a tenth the price of Pro. If you locked in AI budgets based on pricing from six months ago, revisit them. Workloads you deemed too expensive to automate may now be viable.
Your action this week: Audit your organisation’s AI vendor exposure. Map which providers you depend on, what alternatives exist for each use case, and how quickly you could migrate if access were restricted. The businesses that treat AI procurement like any other critical supply chain will be the ones least disrupted when the next shock hits.
This Week’s Policy & Regulation Brief
Trump Administration Blacklists Anthropic, Designates It a National Security Risk
President Trump ordered all federal agencies to cease using Anthropic's technology. Defence Secretary Hegseth designated the company a "supply chain risk to national security" after Anthropic refused unrestricted military use of Claude, including mass domestic surveillance and fully autonomous weapons. Anthropic is challenging the designation in court.
OpenAI Signs Pentagon Deal
OpenAI Signs Pentagon Deal, Then Amends It Under Pressure Hours after Anthropic's blacklisting, OpenAI signed a $200 million classified-network contract. Following a 295% surge in ChatGPT uninstalls and employee open letters, OpenAI added explicit anti-surveillance safeguards. CEO Sam Altman conceded the original announcement was "rushed" and "looked opportunistic."
US Supreme Court Draws the Line on AI Creativity: Fully AI-Generated Works Cannot Be Copyrighted
The US Supreme Court declined to hear Thaler v. Perlmutter, leaving intact a lower-court ruling that copyright protection requires human authorship. Works created entirely by artificial intelligence without meaningful human involvement cannot be copyrighted under US law. The decision effectively settles the issue for now, reinforcing that AI can assist creativity but cannot legally be recognised as the author of a protected work.
OpenAI Closes Record $110 Billion Funding Round
Amazon led with $50 billion, alongside $30 billion each from Nvidia and SoftBank, valuing OpenAI at $730–840 billion. AWS becomes the exclusive third-party cloud for OpenAI's enterprise platform. This cements deep entanglement between the dominant chip maker, the largest cloud provider, and the leading model developer.
March 11 Federal AI Deadline Looming
Three simultaneous deadlines converge next week: the Commerce Department must evaluate state AI laws, the FTC must issue AI deception guidance, and broadband funding tied to state AI regulation takes effect. Colorado's AI Act is directly in the crosshairs.
EU AI Act Enforcement Looms: Companies Face €35M Fines and Criminal Deepfake Penalties from August 2026
The EU AI Act enters full force on 2 August 2026, giving organisations just five months to prepare. Non-compliance could trigger fines of up to €35 million or 7% of global turnover. Some member states are already moving faster: Italy has introduced criminal penalties for the dissemination of deepfakes. For companies deploying AI in Europe, regulatory compliance is rapidly becoming a board-level priority.
AI’s New Cold War: Anthropic Warns Western Models May Be Fueling China’s AI Breakthroughs
Anthropic has warned that Chinese AI labs may be using Western frontier models to help train their own systems, raising concerns about an accelerating global AI arms race. The comments come as reports suggest DeepSeek’s upcoming V4 model may have been trained on restricted Nvidia Blackwell chips, potentially breaching US export controls. Together, the developments highlight growing fears that Western technology is indirectly fuelling China’s next generation of advanced AI models.
DeepSeek Allegedly Trained V4 on Banned Nvidia Blackwell Chips
A senior Trump administration official told Reuters that DeepSeek’s forthcoming V4 model was trained on Nvidia’s most advanced chips in potential violation of US export controls. If confirmed, this would be the most significant export control breach in the AI era.
Model & Platform Updates
OpenAI Rolls Out GPT-5.3 Instant as Default Model with Major Hallucination Reduction
OpenAI has released GPT-5.3 Instant as its new default model, delivering a 26.8% reduction in hallucinations by incorporating web-sourced answers. The update also reduces unnecessary refusals and improves conversational flow, producing more natural interactions. The release signals continued focus on reliability and usability as model providers compete to make AI systems more trustworthy for everyday and enterprise use.
Google Slashes AI Costs with Gemini 3.1 Flash-Lite as Gemini 3 Pro Heads for Shutdown
Google has launched Gemini 3.1 Flash-Lite, a significantly cheaper and faster model designed for high-volume AI workloads. Priced at $0.25 per million input tokens, roughly a tenth the cost of Gemini Pro, it is also 45% faster than the previous generation. At the same time, Google has scheduled Gemini 3 Pro for deprecation on 9 March, giving developers just days to migrate their applications.
Google ships Gemini 3.1 Pro with doubled reasoning performance
Gemini 3.1 Pro more than doubled reasoning performance over 3 Pro on ARC-AGI-2, scoring 77.1%, at unchanged pricing. The rollout went live across the Gemini app, Vertex AI, and Google AI Studio on the same day, combined with Alphabet's 4% stock jump on the enterprise rollout.
Enterprise Agent Programmes Expand
Anthropic unveiled pre-built agentic workflows through Cowork plugins for finance, legal, HR, and engineering. OpenAI announced multi-year Frontier Alliances with Accenture, BCG, McKinsey, and Capgemini. The competition for enterprise AI distribution is fully joined.
DeepSeek V4 Expected Imminently
A trillion-parameter, million-token-context open-weight model with conditional memory and major efficiency gains. If the benchmarks hold up, this positions a Chinese open model closer to frontier-lab capabilities than ever before.
Quick Hits
ChatGPT consumer revolt intensifies - Uninstalls surged 295%, Claude hit number one on the App Store, one-star ratings spiked 775%, and Katy Perry publicly switched to Claude.
Big Tech AI capex hits $650 billion for 2026 - Bridgewater warns the AI boom has entered a “more dangerous phase.”
February VC: $189 billion globally, 90 percent AI - Three companies alone accounted for 83 percent of all venture dollars raised.
OpenAI hints at ChatGPT 5.4 - This is all we know
We work with leadership teams to move from experimentation to execution safely, commercially, and at speed. Talk to us.
