A new draft clause from the U.S. General Services Administration outlines requirements for protecting government data when processed by AI systems. In Europe, recent amendments to the EU AI Act delay the application of obligations for high-risk AI systems. Other topics include California's disclosure of high-risk AI systems, the use of AI to reduce research costs, and the limitations of AI in grant writing.
The U.S. General Services Administration (GSA) released a second draft of its GSAR 552.239-7001 clause. This clause outlines requirements for protecting government data when processed by large language model (LLM) artificial intelligence (AI) systems. The updated draft incorporates feedback from an initial version and applies to contractors handling "Government Data," with some exceptions.
The European Parliament approved amendments to the EU AI Act. These changes delay the application of obligations for high-risk AI systems and simplify regulatory rules. This decision provides businesses more time for compliance, following industry pressure and concerns about readiness.
- California officials disclosed six high-risk AI systems.
- Artificial intelligence reduces the cost and time for conducting research.
- AI does not replace human expertise in grant writing.
Sources
- GSA Seeks Comment on Updated AI Contract Clause: Wiley
The U.S. General Services Administration (GSA) has issued a second draft of its proposed GSAR 552.239-7001 clause, aiming to establish comprehensive requirements for safeguarding government data within large language model (LLM) artificial intelligence (AI) systems. This updated draft reflects significant revisions based on feedback received on the initial version, and public comments are due by August 3, 2026, with a listening session scheduled for July 14, 2026. The clause will apply to contractors whose LLM AI systems process "Government Data," imposing compliance obligations despite…
- EU Approves Delays and Other Amendments to Certain EU AI Act ...
The European Parliament has approved amendments to the EU AI Act, delaying the application of certain obligations for high-risk AI systems and simplifying regulatory rules. This decision follows industry pressure and concerns regarding the timely release of supporting compliance frameworks and guidance. Businesses should primarily view these changes as an extension of time for compliance efforts rather than a significant relaxation of underlying obligations.
Also this week
- economictimes.com
- The best AI grant writing tools for nonprofits in 2026 | Articles | Grant ...
- grantedai.com
- AI and Data Science Research Initiative (2026) | Granted AI
- collibra.com
- HHS details takeaways from sweeping AI request for information ...
- National AI Report – 2026 – Extension Foundation
- AI and the New Future of Work CFP | Spring 2026 - Microsoft Research
- AI creates new opportunities for young researchers and developing countries — expert
- timeshighereducation.com
- soloaiflow.com
- AI Grant Writing in 2026: Benefits, Risks, and Strategy - ScienceDocs
- GSA Proposes Sweeping AI Data Safeguarding Rules for LLM Contractors
- What AI Can and Can't Do for a Grant Writer in 2026 – Grant Writing Made Easy
- California admits using high-risk AI — including systems it failed to ...
Full transcript
How should governments safeguard their data when it's fed into large language models? That's one of the regulatory questions we're covering today on AI in RA, the show that tracks developments in artificial intelligence. We begin with a new draft clause from the U.S. General Services Administration.
The U.S. General Services Administration is trying to formalize how government data is handled by large language models. They've released a second draft of a proposed contract clause to do just that.
So this is about setting the rules of the road for any contractor using AI with government information. It sounds like they're getting specific, down to the contract level.
They are. The clause is GSAR 552.239-7001, and it would apply to any AI system that processes what's defined as "Government Data." There are some exceptions, but that's the main thrust.
And this is a second draft, which means they're reacting to feedback. It's not a final rule being handed down.
Correct. They're still accepting public comments, with a deadline of August 3rd, 2026, and there's a listening session scheduled for July 14th.
Meanwhile, on the other side of the Atlantic, Europe is also making adjustments to its own AI regulations.
The European Parliament just approved amendments to the EU AI Act. The changes push back the start date for obligations tied to high-risk AI systems.
So they're giving businesses more time to prepare. This follows a lot of concern from the industry about being able to comply with the new rules on the original timeline.
Exactly. It's a delay to give them more runway. The underlying obligations themselves haven't been substantially changed, but some of the regulatory rules have been simplified.
Which brings up the question of what's already in use. It's not all theoretical. In California, for example, officials have disclosed they have six high-risk AI systems operating right now.
And at the same time, we're seeing more reports on the practical limits of this technology. One shows that while AI can reduce the cost and time for research, it has its boundaries.
Right. It cannot, for instance, replace human expertise in things like grant writing. The tool can help, but it can't do the whole job.
That covers the latest developments for this week. We will have more updates in our next episode. Until next time, on AI in RA.
From our community
A peer working group for research administrators who are actively building AI into their offices. Some members work alongside IT teams; others are piecing it together on their own. What unites the group is the work of implementation itself: