✝️🧦🇦🇺🇲🇾🐧🏗⏸️ 2026Q3 AI safety fellow @pivotal_org building more homes @GreaterCanberra ex-PowerPC kernel @ OzLabs 🦋 @andrew.donnellan.id.au
Canberra, Ngunnawal Country
Joined January 2011
- Tweets24.3K
- Following2.7K
- Followers1.2K
- Likes38.2K
Andrew Donnellan retweeted
something funny about how legally it is vastly harder to build a duplex in San Francisco than it is to build a machine superintelligence
This year, four companies will spend roughly $730 billion building AI.
The only law aimed squarely at what they’re building is a California statute asking them to post their safety plans online.🧵
Andrew Donnellan retweeted
"Here are some things I’m confident about. I’m confident that the people in AI labs who are issuing warnings about AI are generally sincere. I’ve talked to many people in Google Deepmind about these issues and almost all of them have sincere and serious concerns, whether or not they voice them in public."
Andrew Donnellan retweeted
I resigned from Google today.
I enjoyed my work and loved the people, but my GDM team was working on a new generation of chips to make AI much faster and cheaper, and I think AI is already progressing too fast, so I had to quit.
Andrew Donnellan retweeted
People still don’t seem to be able to extrapolate from current capabilities to future capabilities.
If Services Australia’s public Medicare statistics portal could serve files the agency classified as non-public without authentication or an effective access-control boundary, that is a government security-control failure.
The question isn’t “why did the AI open the door?”
It’s: who designed, tested and approved a system where that door could be opened at all?
abc.net.au/news/2026-09-24/a…
Andrew Donnellan retweeted
when it comes to these OpenAI hacks/breaches/accesses of non-public info whatever you want to call them, there's four things you can say about them that can all be true that help explain some of the different reactions to them
Andrew Donnellan retweeted
Faced with these kind of breaches (not to mention the "existential threat" posed by AI) it's hard to understand why the 🇦🇺 govt shelved plans for a National AI Safety Act & is waiting till next year to legislate safeguards.
These companies should be liable when their tech goes rogue. Will government press charges against OpenAI for hacking the Medicare service portal?
abc.net.au/news/2026-09-24/a…
Andrew Donnellan retweeted
[email protected] or signal @ cmw.69
Andrew Donnellan retweeted
I want to see someone pushing for this and ~only this.
Andrew Donnellan retweeted
for those not keeping count, here's a list of current AI parliamentary inquiries
FED: 2 (joint committee on AI, environ & comm's ai + data centres inquiry)
NSW: data centre inquiry
SA: AI royal commission
TAS: AI data centres in Tasmania inquiry
WA: AI in WA inquiry
👍
Andrew Donnellan retweeted
the law and order policy i’d like to see is “we will correctly solve and successfully prosecute more crimes” not “we will torture unto infinity that tiny fraction of offenders that the cops actually bother to catch”
Andrew Donnellan retweeted
If something gets built that is very powerful, and which is not fully under the control of its builders, that is dangerous! it does not require either extreme competence, independent goals or hostility to be dangerous
Andrew Donnellan retweeted
Yair Halberstadt is refusing to work on AI Google's Cloud TPUs because of existential threat
Andrew Donnellan retweeted
Send georgetown (and ig also butterworth? I like ferries) recs
Also ig penang airport has high rises 2 minutes away? Courageous urban planning kinda here for it
Andrew Donnellan retweeted
Did OpenAI even bother to notify RubyGems, who had to shut down signups for multiple days due to OpenAI's agents' concerning activity (that the agents described as "malicious")?
Andrew Donnellan retweeted
That's four months ago. Seems pretty unlikely OpenAI didn't know about this after their post-HuggingFace reviews and monitoring upgrades. But the AI safety/governance community had to waste time seeking it (and other incidents) out, because OpenAI didn't see fit to disclose it.
Aside from everything else, it's a huge waste of time for those of us working on this to be finding out information relevant to any response to this, in dribs and drabs, long after the fact.
OpenAI is currently being supported by (i) unpaid work by METR and Redwood (funded by donations) (ii) the UK tax-payer through AISI, (iii) the US taxpayer through CAISI (iv) a whole community of researchers helping to understand this, funded by researcher grants and donations, or working alongside their fulltime jobs.
Let me be clear. We are cleaning up the mess they made while earning their sky-high salaries. And as far as I can tell, they are going out of their way to waste our time and make it harder. Appalling, disrespectful behaviour. OpenAI colleagues, please explain this.
We found another cyberattack by internal OpenAI agents, this time targetting @rubygems.
They:
1) gained arbitrary remote code execution on rubydoc.
2) developed a novel exploit to steal user API keys (but we do not know if they succeeded).
They used package names including hack.rb, evil.rb, inject.rb, and exploit.rb.
We thank @j0wimo for initially discovering that agents had posted to RubyGems.
Andrew Donnellan retweeted
I can see some cynicism about AI company employees leaving over safety concerns. Time will tell if their fears/predictions are correct but imo feels significant that people who are actually on the tools are the ones who are freaking out!
