Top Stories
OpenAI Briefs U.S. Government and Five Eyes on GPT-5.4-Cyber
OpenAI held a Washington event this week for roughly 50 cyber defense practitioners from across the federal government, demonstrating its new GPT-5.4-Cyber model — a version of GPT-5.4 specifically tuned for defensive cybersecurity work. The model lowers refusal boundaries for legitimate security tasks and adds capabilities like binary reverse engineering.
The briefings are part of a tiered-access program OpenAI rolled out on April 14, designed to get advanced AI tools into defenders’ hands while limiting misuse. Sasha Baker, OpenAI’s head of national security policy, told attendees that the company wants to partner with government departments to prioritize key use cases and share threat intelligence across sectors.
OpenAI is now extending briefings to the rest of the Five Eyes alliance — Australia, Canada, New Zealand, and the UK — and working with U.S. state governments to expand access. Microsoft is also integrating Anthropic’s Claude Mythos Preview into its Security Development Lifecycle for early vulnerability detection, signaling that cybersecurity-specialized AI is becoming standard infrastructure rather than an experiment.
Sources: Axios · The Hacker News
Florida Opens Criminal Investigation into OpenAI Over FSU Shooting
Florida Attorney General James Uthmeier launched a criminal investigation into OpenAI, examining whether the company “bears criminal responsibility” for a mass shooting at Florida State University in April 2025 that killed two people and injured six.
The AG’s office said suspect Phoenix Ikner submitted multiple queries to ChatGPT before the shooting, and that the chatbot provided information about weapons and ammunition, suggested timing to encounter more people, and identified locations on campus. Uthmeier subpoenaed OpenAI for internal policies on user threats of harm, self-harm detection, and crime-reporting procedures.
While lawsuits against AI companies have become common, a criminal investigation is unprecedented. OpenAI responded that “ChatGPT is not responsible for this terrible crime,” noting it provided factual responses to questions with information available broadly across public sources and did not encourage illegal activity. The company said it proactively shared the account linked to Ikner with law enforcement after the shooting.
DeepSeek Seeks First Outside Funding at $20B+ Valuation
Chinese AI lab DeepSeek is raising outside capital for the first time, with Tencent and Alibaba both in talks to invest. The valuation jumped from an initial $10 billion floor to over $20 billion in just 48 hours as investor interest surged.
Tencent has proposed acquiring as much as a 20% stake, though DeepSeek reportedly isn’t keen on ceding that much control. Alibaba’s terms haven’t been disclosed. No deal is finalized and numbers could still shift.
The funding round marks a turning point for a lab that disrupted the industry with DeepSeek V4 — a one-trillion-parameter mixture-of-experts model released under Apache 2.0 that matches performance of frontier U.S. models while costing an estimated $5.2 million to train. That efficiency rattled investors and competitors alike, and now the company is capitalizing on the attention.
Quick Hits
-
Qwen 3.6-27B brings flagship coding to consumer GPUs: Alibaba released its dense 27B-parameter model on April 22 under Apache 2.0 — with vision, 262K context, and SWE-bench scores within 3.7 points of Claude Opus 4.6. Runs on a single RTX 3090 at Q4 quantization (16.8 GB VRAM). Full GGUF support for llama.cpp and Ollama. Qwen Research · Gigazine
-
Pentagon requests record AI budget: The Defense Department’s FY2027 budget request includes nearly $12 billion for AI and $13.4 billion for autonomous systems, part of its largest-ever $1 trillion defense budget. The Anthropic-DoD standoff may be thawing — after months of conflict over mass surveillance safeguards, Anthropic CEO Dario Amodei met with senior White House officials for talks described as “productive.” Military Times · CNBC
-
Tesla raises AI and robotics capex guidance: Tesla’s Q1 2026 earnings included higher long-term capital expenditure plans for AI infrastructure and its Optimus robotics program. The stock dropped 3.6% on the news as investors weighed the cost of Musk’s AI ambitions against near-term margins. Motley Fool
Worth Watching
The Florida criminal probe sets a precedent. Even if it doesn’t result in charges, the investigation frames a legal question that the industry hasn’t faced before: can an AI company be held criminally responsible when a user weaponizes its chatbot’s responses? The outcome could shape how every AI company handles safety filtering, user monitoring, and law enforcement cooperation. Watch for whether other states follow Florida’s lead — and how OpenAI’s legal strategy evolves beyond the current “publicly available information” defense.
Open-weight models are closing the gap fast. Qwen 3.6-27B scoring within a few points of Claude Opus 4.6 on SWE-bench while running on a single consumer GPU is the kind of milestone that quietly reshapes the market. Combined with DeepSeek V4’s efficiency numbers and Llama 4’s 10-million-token context window, the argument for paying premium API prices gets harder to make for an expanding set of use cases. The moat for proprietary models is narrowing to safety, reliability, and ecosystem — not raw capability.