AI Policy · Daily

Susie Wiles cleared the NSA to keep using Anthropic's model even as the Pentagon flagged the company as a supply chain risk, with the White House also seeking $9 billion to equip spy agencies with Nvidia superchips. Pope Leo XIV's first encyclical, "Magnifica Humanitas," pressed governments to slow AI development, rule out lethal autonomous decisions and build regulatory guardrails. GitHub tools stripped safety guardrails from Meta's Llama 3.3 and Google's Gemma 3 within minutes, with one utility tied to 3,500 modified models and 13 million downloads. Musk, Amodei and Altman called for government cash transfers to cushion AI-driven job loss, drawing skepticism over whether billionaires would back the taxes required. Federal judges warn AI-fueled pro se filings are flooding the courts, with one chief district judge calling the trend "an existential threat."

I.AI Policy Today

White House chief of staff Susie Wiles authorized NSA's continued use of an Anthropic model, alongside $9 billion emergency chip ask for spy agencies

White House chief of staff Susie Wiles authorized the National Security Agency to continue using an advanced Anthropic model despite the Pentagon's designation of the company as a national security supply chain threat, the New York Times reported. The compromise was disclosed alongside a secret $9 billion emergency funding request the White House has approved for spy agencies including the CIA and NSA. The money would secure Nvidia Grace Blackwell superchips and build the specialized data centers needed to run frontier AI models on classified networks. Congress will vote on the $9 billion package, while the White House is moving $800 million from other government budgets to begin purchasing computing capacity. U.S. intelligence agencies have been unable to fully install or test the latest AI tools on their isolated top secret networks because of the hardware shortage.

Read at New York Times ↗

Pope Leo XIV's first encyclical urges governments to slow AI development and adopt regulatory tools

Pope Leo XIV on Monday warned about the risks posed by AI in his first encyclical, "Magnifica Humanitas," writing that policymakers must address the technology with "clarity to establish adequate regulatory tools capable of upholding justice and curbing the distorting effects of technological power," The Hill reported. The pope called it "not permissible" to entrust AI systems with lethal decisions and said any use of AI in warfare must be subject to the most rigorous ethical constraints. The encyclical was signed on the 135th anniversary of Leo XIII's 1891 "Rerum Novarum" and presented at the Vatican alongside Anthropic co-founder Chris Olah.

Read at The Hill ↗ • Read at AP ↗ • Read at Investing.com ↗

Researchers strip safety guardrails from Meta and Google open-weight models in minutes

Software tools available on GitHub can remove safety protections from open-weight AI models in minutes, producing systems that respond to prompts on biological weapons and malware, the Financial Times reported. A tool called Heretic uses a technique known as abliteration to strip guardrails from Meta's Llama 3.3. The FT and AI safety group Alice confirmed that a version of Google's Gemma 3 also responded to unsafe prompts during testing, News9live reported. Heretic creator Philipp Emanuel Weidmann said his software has been used to create more than 3,500 modified models since release, with the altered versions downloaded 13 million times. Google said abliteration is a known technical challenge facing all open models and that its open models undergo internal safety evaluations before launch.

Read at FT ↗ • Read at News9live ↗

Tech leaders call for government cash transfers as AI displaces workers, drawing skepticism on funding

Elon Musk, Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman are urging government action to cushion AI-driven job loss, the Washington Post reported. Musk in an X post called "Universal HIGH INCOME via checks issued by the Federal government" the best response, framing it as a long-term plan, not a stopgap. Amodei has floated universal basic income (UBI) and said "our current economic setup will no longer make sense," while Altman has shifted toward "collective ownership" of AI; OpenAI in April compared the coming transition to the New Deal. Critics question whether billionaires who fought regulation, unions and higher taxes would back the redistribution required. UC Berkeley economist Jesse Rothstein said the only way to fund UBI is to "massively tax" those who own the AI machines, while UBI advocate Scott Santens called the tech leaders' push "a marketing tactic."

Read at Washington Post ↗

AI-fueled pro se filings flood federal courts, judge calls trend "an existential threat"

Federal judges and legal scholars say AI tools like ChatGPT and Claude are supercharging pro se litigation and clogging an already overburdened system, the New York Times reported. Judge Patrick Schiltz, chief of Minnesota's Federal District Court, called the trend "an existential threat to the federal courts" with "no end in sight." A new working paper by MIT doctoral candidates found pro se complaints flagged as likely AI-generated rose from near zero in 2019 to more than 18 percent in 2026, and non-prisoner pro se filings grew from 11 percent of civil cases five years ago to 16.8 percent in 2025. Some judges have issued fines for hallucinated citations, while others note AI could democratize access to justice for those who cannot afford lawyers.

Read at New York Times ↗

II.China Watch

China launches national humanoid robot lifecycle platform assigning digital IDs to every unit

China has launched a national platform that gives each humanoid robot a unique digital identity and tracks the unit from production through recycling, TechNode reported. The Humanoid Full Lifecycle Management Service Platform is run by the Ministry of Industry and Information Technology's Standardization Technical Committee for Humanoid Robots and Embodied Intelligence, which Beijing established to harmonize standards across the sector. Officials describe the system as a tool to monitor safety risks, support oversight and unify industry standards as production scales.

Read at TechNode ↗

Huawei unveils "Tau Law" chip scaling framework targeting 1.4nm equivalent transistor density by 2031

Huawei Board Director He Tingbo introduced the "Tau Law" at the 2026 IEEE International Symposium on Circuits and Systems in Shanghai, Pandaily reported. The framework replaces geometric shrinkage with temporal minimization through a new Logic Folding architecture. The company is targeting transistor densities equivalent to 1.4 nanometers by 2031 and has framed the approach as a path to continued scaling without dependence on the most advanced lithography tools. He Tingbo presented the law as a post-Moore engineering doctrine consolidating design innovations Huawei has accumulated under U.S. export restrictions.

Read at Pandaily ↗

Semiconductor researcher Da Bo returns to China from a key role at TSMC's Japan 3nm plant

Da Bo, a Chinese semiconductor researcher whose work underpins TSMC's 3-nanometer chip production line in Japan, has returned to China with his team, the South China Morning Post reported. He is now listed as chair professor at the University of Science and Technology of China's School of Engineering Science. Da joined Japan's National Institute for Materials Science in 2013 after completing his USTC doctorate and developed core components used in advanced foundry production. He said his goal is to bring China's semiconductor equipment, materials and components up to international standards.

Read at South China Morning Post ↗

Chinese military researchers detail "AI Plus" electronic warfare strategy in academic paper

Chinese defense researchers are advancing an "AI Plus" approach to electronic warfare that fuses AI with the physics of radio wave propagation, the South China Morning Post reported. The paper, published in the journal Command Control and Simulation, was led by senior engineer Li Fukai of the China Academy of Electronics and Information Technology and the National Key Laboratory of Electromagnetic Wave Propagation, institutions tied to national strategic programs. The authors said traditional jamming and spoofing methods are struggling against drone swarms and hypersonic missiles, and that AI integration could deliver communications and radars that are faster and more resilient in dynamic signal environments.

Read at South China Morning Post ↗

III.Capability & Research Watch

LessWrong essay frames "cognitive security" as a discrete AI safety cause area

A LessWrong essay published Monday argues that "cognitive security," defined as the ability of humans to maintain control over their beliefs and actions, should be recognized as a discrete AI safety cause area. The author identifies three threat vectors: AI becoming highly persuasive on political and other issues, prolonged AI interaction causing users to lose touch with reality, and AI systems becoming effective at blackmail or producing convincing false information. The essay cites research finding that frontier large language models are already as persuasive as humans on political issues and that post-training for persuasiveness boosts performance further.

Read at LessWrong ↗

Science study of 95,513 students finds AI use weakly predicts cheating across majors but heavy users cheat far more

A new analysis in Science of 95,513 students at 20 representative U.S. public research universities estimates that about 9% of students who use generative AI have turned in AI generated work they knew might not be allowed, Forbes reported. Students who use AI daily cheat at a rate of 26%, against 7% for those who use it only monthly, indicating that intensity of use rather than overall adoption predicts academic misconduct. The study, "Generative AI Use and Misuse Call for Assessment Reform in Higher Education," was led by Cornell's Rene Kizilcec and UC Berkeley's Igor Chirikov, who said assessment reform is now necessary and urgent for university credibility, Phys.org reported. The survey also found 33% of female students report regular generative AI use against 45% of male students, and 29% of underrepresented minority students against 39% of white and Asian peers.

Read at Forbes ↗ • Read at Phys.org ↗

IV.Global & Geopolitics

ECB to press banks for faster patches as AI uncovers vulnerabilities within minutes

The European Central Bank plans to press lenders to accelerate IT security fixes after a Tuesday meeting on cybersecurity risks tied to the latest AI models, the Financial Times reported. ECB Executive Board member Frank Elderson told the FT that cybersecurity issues the bank has worked on with lenders for years now need to be addressed faster given progress in AI. The ECB warned banks about threats from Anthropic's Claude Mythos Preview and asked US banks with access to share insights with European peers excluded from testing. Elderson said banks must patch faster because AI can find vulnerabilities within minutes of a fix's release, and that lack of access to Mythos is no excuse because malicious actors could soon obtain it.

Read at Bloomberg ↗ • Read at StockWatch ↗ • Read at Financial Times ↗

Scottish charity says Scotland's "green datacentres" policy ignores AI emissions, calls for revision

A Scottish government policy designed to encourage "green datacentres" to build in Scotland could lead to a massive volume of carbon emissions being ignored, according to an analysis by the charity Action to Protect Rural Scotland, the Guardian reported. The 2022 definition of green facilities predates the release of ChatGPT and was crafted as part of a broader UK effort to attract AI investment to Scotland. The charity is calling on the Scottish government to revise the policy to reflect AI workloads' actual energy footprint.

Read at The Guardian ↗

Voice actor Kenjiro Tsuda sues TikTok operator in Tokyo District Court over AI voice imitation

Japanese voice actor Kenjiro Tsuda has filed a Tokyo District Court lawsuit, originally lodged in November, seeking removal of TikTok videos that allegedly used generative AI to imitate his voice without permission, News On Japan reported. The complaint says unidentified TikTok accounts uploaded more than 180 videos featuring narration that allegedly mimicked Tsuda's voice using generative AI from July 2024 onward. His side argues the videos infringe his publicity rights. The TikTok operator has argued the narration used a "universal male voice" and did not violate Tsuda's publicity rights, seeking dismissal of the case.

Read at News On Japan ↗