Martin Radzaj (@martinr) header
Martin Radzaj (@martinr) profile picture

Barrister. AI safety researcher. Safe AI needs lawyers. https://nitter.cf/t.co/Z53sNUCXnR

Joined March 2009
Photo by Martin Radzaj (@martinr)Photo by Martin Radzaj (@martinr)Photo by Martin Radzaj (@martinr)Photo by Martin Radzaj (@martinr)
Martin Radzaj retweeted
Finally having a proper read of Australia's Critical Infrastructure Security Centre's Annual Risk Review for 2026 and the language on AI is surprisingly strong and heavily featured, including recognising power concentration risk.
1
1
2
97
Martin Radzaj retweeted
Governor Newsom’s AI Framework is out! Plenty of reading to do, but I’ve screenshotted the main recommendations in 🧵 gov.ca.gov/2026/11/18/govern…
2
6
40
1,413
Martin Radzaj retweeted
Australia’s PM has just announced new mandatory AI Safety & Security protections which will be legislated, including a new AI regulator, security requirements, risk management obligations, independent evaluation, and mandatory incident notification. youtube.com/live/Qj6hC6KUX-M…
3
1
1
2
1,042
If Australia wanted to build out 10-15 GW of AI compute in the next three years, what would it actually take under the government's constraints of no net costs for consumers and 100% renewable backing? I had great fun trying to work this out with Tim Nelson, one of Australia's sharpest energy economists. The main idea we tested: what if the best place for training data centres is the outback -- far from cities, close to strong sun and wind, and maybe even off the main grid? Watch/listen below or on YouTube, Spotify, Apple Podcasts, etc. Enjoy! Timestamps: (0:00:00) – Introduction. (0:00:42) – Could AI data centres consume half the grid? (0:06:13) – Can we power the build-out with only renewable energy? (0:15:44) – What could bottleneck the build-out? (0:24:36) – How data centres could lower power prices (0:31:58) – What smelters teach us about building renewables quickly (0:35:52) – How would Tim build 10-15 GW of AI compute in 3 years? (0:40:38) – Special compute zones in the outback (1:02:02) – Renewable energy certificates and the time-matching debate (1:12:01) – Nuclear; unintended consequences; Jevons (1:20:19) – Post-interview chat
11
12
7
140
14,892
Martin Radzaj retweeted
Remarkably quick work by the Joint Select Committee on AI in getting some of its early views out yesterday, and I'm very happy to see the quality of its recommendations. Quick initial reactions on the ones that stood out to me: 1/11
5
10
1
39
5,573
I've found that disagreements about AI policy/governance/alignment will often trace back to unstated divergence on base-level facts. My goal here is to establish a shared baseline (model) for thinking about AI. martinsays.substack.com/p/ai…
69
An OpenAI agent got into a Medicare portal in June. Canberra heard about it in September, via a public inbox. Investigators will now mostly work with whatever OpenAI hands over. We wrote about that exact problem after the Hugging Face incident: apartresearch.com/project/ev…
75
Martin Radzaj retweeted
Really excited to share that we have a new piece out in PNAS on legal AI benchmarking from an institutional perspective. Though a huge amount has been written on the technical aspects of benchmarking (e.g., picking metrics and building datasets), relatively less has been written on the institutional design of benchmarking. The piece is about legal AI benchmarking specifically, but it contains two broader ideas which are important to larger conversations about AI governance. (coauthored with @andystruct @christinestsang @chrmanning @JulianNyarko Daniel Ho)
8
11
3
80
8,676
Martin Radzaj retweeted
🧵 How should we think about the field of law and AI? HAI senior fellows Daniel Ho, @JulianNyarko, and @chrmanning and HAI Managing Director @vanessaparli have co-edited a special feature for @PNASNews that explores the appropriate role of AI in law and the role that law has in helping to govern tech. Ft. Cover Artist: Stanford HAI staff, Chris Ellis (1/3) ↘️
8
27
2
73
8,053
Martin Radzaj retweeted
How can AI support public defense? We have worked on this question for the last two years, and have (1) built an AI retrieval tool for the NJ Office of the Public Defender and (2) conducted qualitative interviews with defenders. Our main insights 👇(1/12)
1
8
6
23
5,605
Martin Radzaj retweeted
Also sharing the full diagram of our argument for the bottlenecks between AI capabilities and cheaper legal services
Arms races can channel AI productivity gains into increasing outputs rather than lower costs. Institutions will be very important here to prevent the worst outcomes. Many domains (e.g. academic publishing) are like this and need to be thinking through the possible futures.
2
9
1
70
26,353
Martin Radzaj retweeted
What happens when you roll out custom generative AI to half a country's judges? New paper on Pakistan's courts with @ProfSultanEcon and @gochristoph. In line with wemustactnow.ai -- we provide early empirical evidence on the impacts of transformative generative AI.
4
57
18
188
73,476
Martin Radzaj retweeted
AI-fabricated case citations continue to show up in courts, but can AI also automatically flag hallucinated legal citations to reduce the verification burden? In our new paper (accepted at COLM 2026), we built a benchmark and evaluated several agentic models to find out (1/7)
3
9
2
20
5,228
This is welcome news. Drug law reform is a health and human rights issue that demands urgent attention. The prevailing criminal approach to drugs is doing more harm than good.
BREAKING: The Attorney-General @JaclynSymes has announced that the Government will form a working group with police, health professionals, addiction specialists & youth workers to advise on possible options for a trial of drug treatment notices — the key ask of my Bill. #springst
3
5
I stand in solidarity with Mr Chau, who spends his birthday jailed in Vietnam for peacefully promoting human rights. Please return him to his family: action.amnesty.org.au/act-no… #Write4Rights #W4R2020 #FreeChauVanKham
Caring for 66,455 revellers at risk delivers $7.5m harm-reduction benefit for Sydney. @RohanLeppert thoughts? theconversation.com/caring-f… via @ConversationEDU