Overview
On 8 November 2026, the Associated Press reported, in a story carried by PBS NewsHour, that staff at the US Center for Advancing Innovation and Standards for Super Intelligence (CAISSI) have been told to report informally to an office of the White House. CAISSI sits within the US Department of Commerce and is the federal body that evaluates the capabilities and safety of advanced AI models and autonomous agents. According to the report, the centre has no permanent director, has been through leadership turnover, and has quietly stopped publishing its safety research and model evaluation reports.
The change matters because CAISSI’s published evaluations were one of the few public, government-produced reference points on how frontier AI models perform and where they fail. Outside those reports, much of the publicly available information on a model’s limits comes from the developer’s own documentation. A reporting line into the Executive Office of the President, combined with a halt on publications, raises a reasonable question for anyone who has relied on that work: can the benchmarks still be treated as independent of political direction?
For Australian environmental professionals, this is not abstract US policy. Many of the AI tools used in Australian consultancies, law firms and councils are developed by overseas, largely US, companies. Where consultants use large language models for desktop studies, report drafting or data checking, and where developers, lawyers and council planners rely on those reports, the quality of the underlying tool is part of the evidence chain. The halt on CAISSI publications means one less external reference point that firms can use to check vendor claims about the tools they use.
Key details
The AP report states that CAISSI staff and officials were directed to establish informal reporting lines to the Executive Office of the President. CAISSI was previously known as the US Center for AI Standards and Innovation (CAISI), and before June 2025 operated as the US AI Safety Institute. Its remit includes testing frontier AI models for cybersecurity and biosecurity risks. Civil society groups and technology policy analysts quoted in the reporting are concerned that the new arrangement exposes those safety benchmarks to political interference.
Resourcing is a central part of the story. The report puts CAISSI’s funding at about US$10 million a year, which is roughly one eighth of the budget of the UK’s counterpart body, the AI Security Institute. The centre is operating without a permanent director. A testing programme on that budget, covering models released by some of the largest companies in the world, has limited capacity to run independent evaluations at the pace new model versions are being released.
The transparency issue predates the reporting change. According to the AP report, members of the US Congress had already asked why CAISSI stopped its regular publications assessing the performance and safety bounds of leading commercial models. Those publications gave outside users a way to compare models on a consistent basis without relying solely on vendor claims. The report does not indicate whether or when publication will resume.
The reporting describes the broader policy direction as a lean towards industry self-policing and voluntary commitments. Technical experts quoted in the piece warn that this approach has a structural weakness: commercial developers control how much access outside evaluators get to their models. Caroline Siegel Singh, AI and emerging technology program manager at the Federation of American Scientists, told AP: “These are really powerful institutions that can drastically improve our lives, or the alternative. And the hope is that you can put in nonpartisan or bipartisan technocratic experts to enact policy in a way that best serves the American public.”
Australian context
Australia does not have a statutory AI regulator, and AI governance for businesses here rests largely on voluntary frameworks and existing law. The main reference points are Australia’s AI Ethics Principles (2019), the Voluntary AI Safety Standard released by the Department of Industry, Science and Resources in September 2024, and AS ISO/IEC 42001:2023, the Standards Australia adoption of the international standard for AI management systems. The Voluntary AI Safety Standard sets out ten guardrails, including testing and monitoring of AI systems, meaningful human oversight, transparency across the AI supply chain, and keeping records that allow third parties to assess compliance. The Commonwealth also announced in late 2025 that it would establish an Australian AI Safety Institute, but Australian practitioners should not assume it will publish model-level evaluations of the tools they use day to day.
For professional services firms, the practical effect of the US change is that the burden of demonstrating a tool is fit for purpose sits with the firm using it. That has always been the legal position. Professional negligence, the Australian Consumer Law, the Privacy Act 1988 and the terms of professional indemnity policies all attach to the person or company that delivers the advice, not the software vendor. Overseas frameworks such as the EU AI Act (Regulation (EU) 2024/1689), the OECD AI Principles and guidance from the UK AI Security Institute remain useful benchmarks, and are now more relevant to Australian firms than waiting for US federal technical baselines.
In contaminated land practice, the reliance chain is long and formal. Data quality objectives under Schedule B2 of the National Environment Protection (Assessment of Site Contamination) Measure 1999, as amended in 2013 (NEPM 2013), require that data and interpretation be defensible and traceable. Reports are reviewed by accredited site auditors under the Contaminated Land Management Act 1997 (NSW), with equivalent auditor schemes operating in other states, and may be relied on by regulators, planning authorities, purchasers and financiers. An auditor or regulator reviewing a site assessment will expect the consultant to explain how each conclusion was reached. If an AI tool was used to summarise historical records, screen laboratory results against investigation levels or draft the interpretation, the consultant needs to be able to show what the tool did and how its output was checked.
What this means for practice
The loss of published US government evaluations does not change any Australian obligation, but it does leave firms with less outside material to point to when asked why a tool was considered reliable. Firms using AI in technical work can reduce that exposure by recording which tools and versions were used and for what tasks, checking AI-generated outputs against primary data and source documents, keeping a record of human review, and reading vendor terms for limits on accuracy and data handling. Aligning internal practice with the Voluntary AI Safety Standard guardrails and AS ISO/IEC 42001:2023 gives firms a documented basis for those decisions that does not depend on any overseas agency continuing to publish.
References and related sources
- Primary source: www.pbs.org
- digg.com
- ground.news
How iEnvi can help
iEnvi integrates technology and data-driven approaches into environmental consulting. We monitor AI and technology developments that affect how environmental professionals deliver services to clients.
This is an iEnvi Machete news summary. Prepared by iEnvi to summarise the source article for environmental professionals tracking AI, data, and technology developments that affect consulting and project delivery.
Published: 09 Nov 2026
Need advice on this topic? Speak to an iEnvi expert at info@ienvi.com.au or 1300 043 684, or contact us online.
Need advice on this issue? iEnvi provides practical, senior-led environmental consulting across contaminated land, remediation, ecology and environmental risk.