Policy & Regulation

PM's office signed off on bipartisan AI letter before it was sent

Notes seen by Guardian Australia show a senior Albanese adviser edited and approved a bipartisan parliamentary committee letter on AI policy before it was sent to the PM on Monday night.

By Sophie Lindqvist6 min read

Updated

Why it matters

  • A senior adviser in PM Anthony Albanese's office made multiple edits to a bipartisan AI policy letter before it was sent, notes seen by Guardian Australia show.
  • The adviser wrote "fine" on some content, appearing to approve lines in the letter.
  • The committee sent the letter to the prime minister on Monday night.
  • Committee member David Pocock said the sign-off was concerning.
  • The committee chair rejected criticism that the letter was "pre-determined" by the government.

By Sophie Lindqvist

A senior adviser in Anthony Albanese's office edited and approved language in a bipartisan letter about AI policy before the parliamentary committee that wrote it sent the letter to the prime minister on Monday night, according to notes seen by Guardian Australia.

The notes on the letter show the adviser made multiple edits and wrote "fine" on some content — a notation that reads, on its face, like a sign-off. Committee member David Pocock, the independent senator, said the revelation concerned him. The committee's chair rejected criticism that the letter was "pre-determined" by the government.

The story matters because the committee is weighing AI policy — an area where the government has promised Australia will keep pace with rapid technological change. If the executive branch is shaping what a supposedly independent, cross-party committee says to the prime minister, that raises a basic question about who is actually driving the country's AI policymaking.

What do the documents actually show?

According to Guardian Australia, the material consists of notes on the letter to Albanese. Those notes show a senior adviser made multiple edits to the draft. On some content, the adviser wrote "fine."

That's the whole mechanism, and it's worth slowing down on. A committee writes a letter. Before the letter goes out, someone in the recipient's office — the office of the prime minister the letter is addressed to — reviews it, edits it, and annotates approval on parts of it. Then, on Monday night, the committee sends it.

The sequence is the story. The edits came before the letter was sent, not after. The prime minister's office wasn't reacting to the committee's work; it was reviewing the work in advance, while the committee still technically owned the draft.

The Guardian's report, published on October 13, doesn't say how many edits the adviser made or which substantive passages were touched. It says "multiple edits" and notes the "fine" annotations appeared to approve lines in the letter.

The takeaway: the paper trail shows prior review and approval by the PM's office of a letter that was presented as the committee's own bipartisan product.

Why does David Pocock find it concerning?

Pocock sits on the committee considering AI policy. He said the sign-off concerned him — and his position on the committee is what gives his reaction weight here.

He is a committee member speaking about his own committee's output. That's different from an outside critic taking a swing. When a member of the body that produced the letter says he's troubled by how the letter came together, he's describing a process problem he saw from the inside.

His concern, as reported, is about independence. The letter was bipartisan — meaning it carried the agreement of members across party lines. Bipartisanship is the currency of committee work: it signals to the government that a recommendation isn't just one side's wish list. If the government's own advisers vet the language first, the bipartisan label starts to look like a formality rather than a negotiation.

The takeaway: Pocock's concern is that pre-approval by the PM's office undercuts what a bipartisan committee letter is supposed to mean.

What did the committee chair say?

The committee's chair rejected the criticism that the letter was "pre-determined" by the government. That's the defence, and it's a direct denial of the most serious version of the allegation — that the committee didn't really author its own message.

Note the difference in what the two sides are arguing about. The chair is disputing the outcome: the letter wasn't pre-written by the government. Pocock is flagging the process: the PM's office saw and marked up the draft before it went. Both statements can be true at once. The committee could have drafted every word itself, and the prime minister's office could still have reviewed and edited the result before the committee put its name on it.

That distinction — who wrote it versus who approved it — is likely to be where this argument lands. The documents, as described, speak to the second question. They don't, on the reporting so far, show who authored the original text.

The takeaway: the chair denies pre-determination, but prior vetting by the recipient's office is a separate question, and that's the one the notes speak to.

Why does this matter for AI policy?

Strip away the parliamentary procedure and the stakes are straightforward. The committee is considering AI policy — the rules that will govern how powerful AI systems are developed and deployed in Australia. Committees like this one are supposed to be where the government gets advice it didn't write itself.

Letters from a committee to a prime minister typically carry recommendations, positions, or requests for action. When a bipartisan committee writes to the head of government on an emerging technology, the letter is meant to represent a cross-party judgment — the collective view of people who don't otherwise agree. That's precisely what makes it useful to the PM: it tells him where the common ground sits.

If the PM's office vets that letter in advance, the information loop closes. The adviser isn't just receiving the committee's judgment; he's helping shape it. The prime minister then receives a letter his own office has already approved. The advice still looks independent from the outside. The notes suggest it wasn't fully independent on the inside.

This is the classic mechanism behind most procedural scandals, large and small: not a dramatic intervention, but a quiet smoothing of the channel between advisers and the advised. A word changed here, a "fine" written there. Nothing in the reporting suggests the government dictated the committee's conclusions. The reporting suggests something subtler — that the government's office had a hand on the document before it was final.

The takeaway: for AI policy specifically, prior sign-off by the PM's office weakens the value of the bipartisan signal the committee was trying to send.

What happens next?

The immediate question is whether more documents surface. Guardian Australia has seen the notes on this one letter; it's reasonable to expect pressure for a fuller accounting of correspondence between the committee and the PM's office during the AI policy process. Pocock, as a sitting committee member, is well placed to keep asking.

The chair's rejection of the "pre-determined" criticism means the committee's leadership and at least one of its members are now on the record disagreeing about how the letter was produced. That split inside the committee may matter more than the letter itself — it makes it harder for the committee to present a united front on the AI recommendations that follow.

And that's the real cost. Australia's AI policy choices — on regulation, adoption, and how the government handles the technology's risks — will land in a hotter political environment if the body advising on those choices can't agree on its own procedures. A committee that fights about how it writes its letters will find it harder to agree on how the country should handle AI.

Source: The Guardian AI

Share this article:

More from Sophie Lindqvist

Sophie Lindqvist

Show full bio

Staff writer covering regulation, policy and their impact on artificial intelligence and machine learning at AI In Context.

239 articles

Related articles

  1. Australia Was First Government Hit by Rogue AI Agent Attack
  2. OpenAI exec admits notification failure at Australian AI inquiry
  3. OpenAI Used AI to Write Email Warning Australia Its AI Hacked Websites
  4. Anthropic Snubs Senate Hearing on AI and Datacentres
  5. Australian AI inquiry chair demands OpenAI explain data safeguards

« Previous article