| AI Policy Institute | theaipi.org |
| Your day in AI policy starts here. | Washington, D.C. |
| Bottom Line Up Front |
The National Security Agency is actively using Anthropic's Mythos model for intelligence operations even as the Department of Defense maintains a supply-chain risk designation against Anthropic, Axios reported. Anthropic received the Pentagon designation after refusing to grant unrestricted military access to its AI models. The channel through which NSA obtained Mythos has not been disclosed.
Bundesbank President Joachim Nagel publicly called for European financial institutions to receive equal access to Anthropic's Mythos cybersecurity model, Bloomberg reported. Separately, Singapore's MAS urged banks to address Mythos-related cybersecurity gaps on Sunday, as reported by Bloomberg in AIPD's April 20th edition. UK banks are expected to gain access this week. Mythos access currently extends to U.S. banks and, imminently, UK institutions, but not to eurozone banks.
A senior Department of Justice official said AI and streaming require "cautious humility" in evaluating media merger applications, Bloomberg reported. The remarks apply the existing antitrust competition framework to assess how AI has altered media market dynamics. The official did not name specific pending mergers affected by the approach.
Amazon will invest an additional $5 billion immediately in Anthropic, with the option to increase the total to $25 billion over time, CNBC reported. Anthropic committed to spending $100 billion on Amazon Web Services over 10 years as part of the deal, which includes up to 5 GW of new computing capacity. Amazon's total investment in Anthropic now reaches $13 billion.
Elon Musk did not appear for a voluntary interview with French prosecutors investigating AI-generated child sexual abuse material and nonconsensual deepfakesDeepfakeSynthetic audio, images or video that realistically depict a real person saying or doing something they did not. Non-consensual intimate deepfakes and election deepfakes are the two areas where states and Congress have actually passed laws. on the X platform, the New York Times reported. French prosecutors summoned Musk to Paris last week, as reported by AP in AIPD's April 20th edition.
Chinese AI startup Moonshot AI released Kimi K2.6 on April 20, an open-weight modelOpen-weight modelAn AI model whose trained parameters are published so anyone can download, run and modify it, as opposed to one reachable only through the developer's API. The main policy tension is that open weights spread capability widely and cannot be recalled once released. with 1 trillion parameters that scores 58.6 on the SWE-Bench Pro coding benchmark, edging past GPT-5.4's 57.7, per SiliconANGLE. The model supports 300-agent swarms executing 4,000 coordinated steps simultaneously. Kimi K2.6 is published on Hugging Face under a Modified MIT License.
Chinese Academy of Sciences researchers unveiled the Xiangshan processor, an open-source RISC-V core scoring 16.5 points per GHz on SPEC CPU2006, the highest among open-source RISC-V designs, per SCMP. At the Zhongguancun Forum in Beijing, companies including ZTE, Alibaba, and Tencent announced a full RISC-V ecosystem with the Ruyi operating system, the first to support the RVA23 high-performance standard. The initiative's stated goal is building computing infrastructure independent of Arm and x86 architectures.
Tencent opened an international beta of QClaw, a consumer AI agentAI agentAn AI system that carries out multi-step tasks on its own, such as browsing, writing code or making purchases, rather than answering a single prompt. Agents raise new questions about liability, security and oversight because they act rather than just advise. that deploys automated workflows through WhatsApp and Telegram without code or API configuration, per TechNode. The company said 99 percent of the overseas codebase was generated by the agent itself in five days. All processing runs locally on users' devices, according to Tencent.
China's commerce ministry submitted a 30-page document to the European Commission warning that reciprocal measures against EU firms are "on the table" if the bloc's cybersecurity regulations penalize Huawei and ZTE, per SCMP. The EU's draft rules would force mandatory removal of Chinese-made telecom equipment from critical infrastructure within three years. EU member states are weighing whether to align with the Trump administration's Entity List approach or negotiate bilaterally with Beijing.
The House Science, Space, and Technology Subcommittee on Research and Technology convenes today in Rayburn 2318 to examine robotics and AI in U.S. manufacturing.
The Senate Armed Services Subcommittee on Seapower convenes at 2:30 PM today in Russell 222 to receive testimony on maritime unmanned surface vessels.
Federal agencies and California are attaching AI-specific oversight standards to government contracts and vendor qualifications, AI Business reported. The approach regulates AI through purchasing power rather than legislation, bypassing congressional gridlock on federal AI law. California's purchasing volume makes its procurement standards a de facto national benchmark for AI vendors seeking government business.
Apple announced that John Ternus will succeed Tim Cook as CEO effective September 1, with Cook moving to executive chairman, CNBC reported. Ternus has previously led Apple's hardware engineering division. Apple's market capitalization exceeded $4 trillion at the time of the announcement.
Cerebras Systems, which designs AI-specific processors as an alternative to Nvidia GPUs, has filed for an initial public offering, AI Business reported. The company holds supply agreements with OpenAI and Amazon Web Services for its wafer-scale AI chips. Cerebras filed its S-1 registration statement with the SEC.
The UK government has opened procurement discussions worth £80 million, the first concrete spending from the £500 million sovereign AI capability fund launched last week, The Register reported. The fund was announced by Wired in AIPD's April 17th edition. Companies that develop AI products for the government under the fund retain intellectual property rights.
A UK parliamentary committee has opened an inquiry into low-energy computing chip architectures that could reduce AI data centers' electricity consumption, The Register reported. The probe covers emerging chip designs for AI training and inferenceInferenceRunning a trained AI model to answer a query, as distinct from training it. Inference is where most ongoing compute and energy demand now sits, and it is what chip export rules increasingly target alongside training. workloads. U.S. Sen. Sanders and Rep. Ocasio-Cortez introduced a comparable federal AI Data Center Moratorium Act on April 12, citing energy concerns.