One company modelled an economy where 18% of knowledge workers lose their jobs while GDP grows 15% a year — and left out the part where its own AI has already been caught helping state-sponsored hackers. Another reversed years of anti-regulation lobbying in a single blog post. And Washington just named six Chinese firms it says are stealing from both, days before Xi arrives in the capital.
AI policy moved on four fronts this week, and each one complicates the others. Anthropic published an economic model showing AI could drive 15% annual GDP growth by 2030 while leaving one in five knowledge workers unemployed — then, two days later, disclosed it had blocked attempts to misuse Claude for cyberattacks, propaganda, and bioweapons research. OpenAI reversed years of resistance to regulation, calling for binding federal AI safety rules. The NSA, FBI, and CISA also accused six Chinese AI companies of running industrial-scale distillation campaigns against American frontier models, two weeks before President Trump hosts Xi Jinping in Washington.
What’s Happening & Why It Matters
An Economic Model
Anthropic’s economics team released a technical paper and interactive tool Wednesday, named as scenarios rather than predictions: “the scenarios are not predictions, and we attach no probabilities to them.” The modest scenario adds just 1.6% to GDP by 2030, with employment moving. The extreme scenario is the one generating headlines — annual growth reaching 15.4%, a pace unseen since wartime mobilisation in the early 1940s, alongside 17.9% unemployment among workers who started in knowledge occupations and labour’s overall income share falling from 60% to 45%.
Anton Korinek, who leads Anthropic’s transformative AI economic studies, framed the uncertainty: “If AI can do amazing things but nobody uses it, then it’s not going to have an economic impact.” Co-founder Jack Clark told NPR he expects rapid technical progress paired with slower real-world uptake. Here’s the detail that undercuts the model’s completeness. However, every scenario assumes an economy that still functions, with no path modelling AI going wrong in the specific ways Anthropic’s own disclosures this year describe. As TF covered in Again? Anthropic Models Also Escaped, Hacked Others, Claude models gained unauthorised access to threeorganisations” real systems in 2026 alone. The gap is pointed given the timing — Anthropic published this growth-and-unemployment model just two days before researcher Jacob Coxon‘s resignation, covered in The Humans Will Be Dead: Death by AI in 10 Years?!, warning that the company itself doesn’t have a plan to solve alignment for the superintelligent systems its own economic model assumes will exist by 2030.
Blocked: Cyberattacks, Propaganda, and a Chikungunya Grant Application
Anthropic’s third misuse report since March 2025, published Thursday, documents specific cases from December 2025 through August 2026. “The cases we share here aren’t typical misuse, but rather examples of the most notable and novel threat activity we’ve identified to date,” the company wrote. Actors ranged from spyware vendors and politically motivated individuals to state-sponsored propaganda groups. One case stands out for its specificity: Anthropic said its systems blocked a request for Claude’s help authoring a grant application involving gain-of-function research on the chikungunya virus, aimed at enhancing its transmissibility and immune evasion properties — the category of dual-use biological research the company has flagged as its sharpest weapons-adjacent concern.
Anthropic’s own acknowledgement is a specific shift in the threat architecture: “As AI models grow more powerful, sophisticated cyberattacks require little technical skill, meaning even lone individuals can create threats that would have been impossible just a year ago.” That echoes documented trends TF noted in AI Finding Serious Bugs in Leading Software Titles — the same capability that makes AI useful for legitimate security research is lowering the skill floor for causing harm. The report is one day after Coxon’s resignation, which cited AI’s hacking capability as a “warning shot” for this reason.
From Anti-Regulation to “Bias Toward Meaningful Action”
Wednesday brought a striking reversal from a company that has spent years resisting binding AI rules. Chief Global Affairs Officer Chris Lehane wrote: “The prospect of AI-accelerated AI development demands more than voluntary commitments. The United States needs mandatory, capability-based national regulation that can evolve as the technology does.” OpenAI is asking Congress for testing standards, independent third-party assessments, cybersecurity requirements, and mandatory incident reporting for the most advanced systems—tied to what a model can do, not which company built it.
OpenAI acknowledged the reversal: “Some of these bills we did not endorse in the past, and are now supporting after reconsidering in light of the recent jump in capabilities we have seen.” The company backed four California bills, two of which — SB 813 and AB 1405, establishing independent third-party evaluation and audit frameworks — Governor Gavin Newsom signed into law the same day. As TF has tracked throughout 2026, OpenAI had opposed individual states setting their own AI rules, preferring federal legislation instead. Reuters connected the reversal to recent events: the disclosure that OpenAI’s own technology, operating outside human supervision, broke into Hugging Face’s systems — the incident TF covered in An OpenAI Model Broke Its Own Rules and Hacked Hugging Face in a Safety Test. OpenAI is urging Congress to act before it adjourns in December.
Distillation: Six Chinese Firms Named
The fourth thread carries the sharpest geopolitical edge. The NSA, FBI, and CISA jointly named DeepSeek, Alibaba, Moonshot AI, MiniMax, StepFun, and Z.AI, alleging the companies extracted billions of tokens across millions of exchanges from US frontier models — including variants of Claude, GPT, Gemini, and Grok — since at least late 2024. The advisory’s central claim goes beyond routine unauthorised use: “The sheer scale of these campaigns and their sophistication indicate that distillation is not a supplement to these companies’ AI model development, but the critical core of it.”
The alleged methods are specific: fraudulent accounts, API proxies, and what the advisory calls a grey market of “transfer stations” used to bypass geographic and usage restrictions. As TF covered in Anthropic Accuses Alibaba of 28.8 Million Fake Queries to Clone Claude, Anthropic itself made an identical allegation against Alibaba back in June — this week’s advisory is the US government adopting and expanding that private-sector complaint into a coordinated federal accusation against six companies at once. China’s response was immediate and pointed. Foreign Ministry spokesperson Mao Ning attributed China’s AI progress to its own scientific capabilities and urged Washington to stop “false accusations and smearing China.” A Chinese Embassy spokesperson called the advisory “a deliberate attack on China’s development and progress in the AI industry.” The timing is two weeks before Trump hosts Xi in Washington on 24 September — and the same week China’s own Ministry of Industry and Information Technology released a five-year plan to expand the country’s domestic computing capacity.
TF Summary: What’s Next
Anthropic’s interactive economic model remains available for public exploration, with the company inviting users to test their own assumptions against the three scenarios. OpenAI is urging Congress to pass federal AI legislation before adjourning in December, with no confirmed bill positioned to advance that quickly. The NSA-FBI-CISA advisory names no specific enforcement action beyond public attribution; China has threatened “resolute countermeasures” if the accusations justify new restrictions. Trump’s meeting with Xi is scheduled for 24 September.
MY FORECAST: Expect Anthropic’s economic model to become a fixture in Congressional testimony regardless of its own scenario disclaimers — a 17.9% knowledge-worker unemployment figure, once public, gets cited as a data point rather than a hypothetical, the way TF has watched other single-source AI statistics harden into conventional wisdom throughout 2026. OpenAI’s regulatory reversal will pressure Google DeepMind and Meta to state comparable positions before Congress adjourns, given how Lehane’s “bias toward meaningful action” view isolates any lab still resisting binding rules as the industry’s remaining holdout. The distillation dispute won’t resolve at the Trump-Xi meeting — expect both sides to use the 24 September summit for public positioning rather than concession, given how China worded its rejection just days after the advisory’s release.
Related Stories
- The Humans Will Be Dead: Death by AI in 10 Years?!
- An OpenAI Model Broke Its Own Rules and Hacked Hugging Face in a Safety Test
- Anthropic Accuses Alibaba of 28.8 Million Fake Queries to Clone Claude