Who can access the most powerful AI, and who oversees its use?
Provisional · Evidence through October 31, 8:30 a.m. MDT · Updated November 1, 2026
Direction of power
Leaning concentration
Window
Narrows overall; evaluation access Opens locally
Confidence
Moderate
Movement since September
More external evaluation access, while strategic access and enforcement remain selective
October assessment
The cumulative position remains provider- and state-controlled frontier access. October’s PAIPA designations extend an AI trade-secrets enforcement action to companies and financing entities. Earlier trusted-user, export-control and research-security arrangements remain the starting position; a quiet day does not erase them.
METR’s October 15 embedded arrangement at Anthropic adds a meaningful counterweight: evaluators receive deep internal access and specified publication rights. The October 16 clarification matters: Anthropic ultimately decides whether material falls within categories it can require METR to redact. Access is not equivalent to unconditional publication, regulatory intervention authority or assured continuity.
The proposed UK Biosecurity Lab combines public funding with supplier credits and access commitments. It may build scientific capacity, but the proposed independent nonprofit’s governance and contractual rights remain open. Internal containment tests show the technical resources oversight can require; they do not establish public control.
What would change this judgment
Enforceable testing and publication rights, demonstrated evaluator independence, dispute experience, sustained public technical capacity, or a materially new access restriction. The earlier UK/US access sequence and the Accenture comparison remain unverified.
Supporting scored events
Each action retains one event and score across lanes. Dates, jurisdictional limits and unresolved verification remain explicit.
2026-10-08 · Dispersion +1 · Secondary lane; primary Lane 1
On October 8, the UK Information Commissioner’s Office (ICO) reported privacy changes secured through its supervision of ten major AI developers: Amazon, Anthropic, Apple, Cohere, DeepSeek, Google, Meta, Microsoft, OpenAI and Stability AI. These are companies developing many models, not ten individual large language models. Some changes are completed; others remain commitments. The work applies existing UK data-protection duties to model training and use, including explaining whose personal data is used, identifying a lawful basis and helping people exercise access and objection rights.
For example, the supervision account documents changes to privacy notices and explanations by Apple, Cohere and OpenAI, and reviews of the justification for using personal data by other developers. On the same day, the ICO opened a public call for evidence on AI agents, which can take actions for users. It closes November 20 and asks about transparency, accountability, security and automated decisions before future guidance and a statutory code. The consultation is open now; the future code is not yet adopted. The underlying data-protection law is compulsory even where firms cooperate without a new enforcement order.
2026-10-15; October 16 redaction clarification · Dispersion +1 · Primary lane
Anthropic announced on October 15 that six METR evaluators received company hardware, office space and internal-system access broadly comparable to its own risk-assessment teams, subject to specified exclusions.
Anthropic added an October 16 clarification to its October 15 announcement about six METR employees working inside the company. It now explicitly says Anthropic ultimately decides whether information falls within categories it can require METR to redact. METR retains specified rights to describe the access and information it received. The arrangement still permits publication without Anthropic’s general approval; it does not give METR final authority over confidentiality classifications. Current primary account.
2026-10-19 · Dispersion +1 · Secondary lane; primary Lane 1
An October 19 federal order, reported October 20, allows the Environmental Protection Agency to join the NAACP's Clean Air Act case over xAI's gas-fired power plant in Southaven, Mississippi, while rejecting the federal government's claimed power to join and unilaterally dismiss it.
The NAACP and its Mississippi conference sued in April over alleged operation without required permits. Judge Debra Brown held that the statute permits EPA intervention but does not grant the dismissal power the administration asserted. The 22-page order establishes an actual ruling. It neither decides the pollution allegations nor orders the plant closed.
This preserves an independent enforcement route for affected communities when the executive branch prefers that their case end. That is a concrete countervailing-power mechanism, even though the underlying environmental outcome remains unresolved.
On October 19, the U.S. State Department designated five China-based entities under the Protecting American Intellectual Property Act in connection with theft of Google AI-related trade secrets. The targets include Shanghai Zhisuan Technology, MiraclePlus and three associated investment or management entities. The department’s statement and fact sheet, reproduced by Forth, establish actual designations rather than a proposal.
The fact sheet specifies sanctions under Section 2(b)(1)(A), (D), (H), (I) and (J). These concern property within U.S. jurisdiction or U.S. persons’ control, loans or credits from U.S. financial institutions totaling more than $10 million in a twelve-month period, with the statute’s humanitarian-relief qualification, foreign-exchange and banking transactions subject to U.S. jurisdiction, and U.S. investment in the entities’ equity or debt, subject to statutory scope and exceptions. This reaches organizations and financing relationships beyond the prosecution of one engineer; it is not a general ban on Chinese AI.
The power shift is the federal government’s concrete exercise of national-security authority to narrow the targeted AI-related network’s access to assets and finance within U.S. reach. That supports a modest concentration assessment even if the security rationale is defensible. It does not establish increased Google market share or a measured decline in Chinese AI capability.