Harry Coppock

201 posts

Harry Coppock

@HarryCoppock

No. 10 Downing Street Innovation Fellow | Research Scientist at AISI | Visiting Lecturer at Imperial College London Working on AI Evaluation and AI for Medicine

London Katılım Mart 2021

342 Takip Edilen209 Takipçiler

Harry Coppock retweetledi

AI Security Institute@AISecurityInst·1d

🔓 Can today’s AI agents escape sandbox environments? Using our new benchmark, SandboxEscapeBench, we find that frontier models can reliably exploit common vulnerabilities - and that breakout capability improves as model size and inference compute increase. Read more ⬇️

English

155

12K

Harry Coppock retweetledi

David@DavidDAfrica·9 Mar

Can LLMs tell when their conversation history has been tampered with? We tested 14 models across thousands of conversations to find out. Some new work from UK AISI 🧵

English

164

15.2K

Harry Coppock retweetledi

AI Security Institute@AISecurityInst·5 Mar

AI cyber capabilities are improving rapidly, but are evaluations keeping pace? Alongside @Irregular, we found that recent models can productively use 10-50x larger token budgets than typical evaluation settings allow, with key security implications🧵

English

19.1K

Harry Coppock retweetledi

AI Security Institute@AISecurityInst·25 Şub

How can we make sense of the vast transcripts generated during agentic evaluations and multi-turn conversations? Together with @meridianlabs_ai, we built Inspect Scout, an open-source transcript analysis tool, and distilled best practices into a step-by-step pipeline🧵

English

3.3K

Harry Coppock retweetledi

AI Security Institute@AISecurityInst·17 Şub

AI companies deploy safeguards that are robust to thousands of hours of human attacks. Today, we share Boundary Point Jailbreaking (BPJ), the first fully automated attack to break the safeguards of leading AI models🧵 (1/8)

English

150

34.6K

Harry Coppock retweetledi

Rishi Sunak@RishiSunak·10 Şub

One of the reasons we created the UK AISI when I was Prime Minister was to have exactly this kind of independent red-team capability. Great to see @soundboy and the team continuing their valuable work so we can unlock the benefits of AI while keeping people safe.

Xander Davies@alxndrdavies

UK AISI's Red Team tested both OpenAI + Anthropic's models released today! We jailbroke GPT-5.3-Codex (and the conversation monitor) in 10 hours & conducted an alignment audit on Opus 4.6. 🧵

English

715

130.8K

Harry Coppock retweetledi

Geoffrey Irving@geoffreyirving·18 Ara

New report on trends in AISI's evaluations of frontier AI models over the past two years. A lot of AI discourse focuses on viral moments, but it is important to zoom out to the less flashy trend: AI models are steadily growing in capabilities, including dual-use capabilities.

AI Security Institute@AISecurityInst

📈 Today, we’re releasing our first Frontier AI Trends Report: evaluation results on 30+ frontier models from the past two years, showing rapid progress in chemistry and biology, cyber capabilities, autonomy, and more. ▶️Read now: aisi.gov.uk/frontier-ai-tr…

English

8.3K

Harry Coppock retweetledi

Cozmin Ududec@CUdudec·10 Eki

Very excited that this systematic analysis is out! We found a bunch of failure modes, as well as interesting and surprising behaviours. Theres a lot more insight we can get from looking carefully at how models are solving evaluation tasks!

AI Security Institute@AISecurityInst

Measuring how often an AI agent succeeds at a task can help us assess its capabilities – but it doesn’t tell the whole story. We’ve been experimenting with transcript analysis to better understand not just how often agents succeed, but why they fail 🧵

English

715

Harry Coppock retweetledi

Anthropic@AnthropicAI·9 Eki

New research with the UK @AISecurityInst and the @turinginst: We found that just a few malicious documents can produce vulnerabilities in an LLM—regardless of the size of the model or its training data. Data-poisoning attacks might be more practical than previously believed.

English

248

1.6K

532.1K

Harry Coppock retweetledi

Robert Kirk@_robertkirk·30 Eyl

We at @AISecurityInst recently did our first pre-deployment 𝗮𝗹𝗶𝗴𝗻𝗺𝗲𝗻𝘁 evaluation of @AnthropicAI's Claude Sonnet 4.5! This was a first attempt – and we plan to work on this more! – but we still found some interesting results, and some learnings for next time 🧵

English

8.2K

Harry Coppock retweetledi

Xander Davies@alxndrdavies·13 Eyl

Excited to share details on two of our longest running and most effective safeguard collaborations, one with Anthropic and one with OpenAI. We've identified—and they've patched—a large number of vulnerabilities and together strengthened their safeguards. 🧵 1/6

English

297

60.6K

Harry Coppock retweetledi

AI Security Institute@AISecurityInst·8 Eyl

🔎 People are increasingly using chatbots to seek out new information, raising concerns about how they could misinform voters or distort public opinion. But how is AI actually influencing real-world political beliefs? Our new study explores this question 👇

English

2.8K

Harry Coppock retweetledi

Robert Kirk@_robertkirk·29 Ağu

Since I started working on safeguards, we've seen substantial progress in defending certain hosted models, but less progress in measuring & managing misuse risks from open weight models. Three directions I want explored more, drawn from our @AISecurityInst post today 🧵

English

2.2K

Harry Coppock@HarryCoppock·15 Ağu

This is great news for the UK. Having worked with Jade over the past 2 years, setting up @AISecurityInst, I am confident that there are very few, if any, who are better placed to take on this role.

Matt Clifford@matthewclifford

Absolutely delighted about this - major upgrade on the last AI adviser! Jade brings a tonne of experience in frontier labs, VC and government and will do an amazing job of ensuring the UK is an AI winner. Excellent news.

English

303

Harry Coppock retweetledi

AI Security Institute@AISecurityInst·12 Ağu

How can open-weight Large Language Models be safeguarded against malicious uses? In our new paper with @AiEleuther, we find that removing harmful data before training can be over 10x more effective at resisting adversarial fine-tuning than defences added after training 🧵

English

219

35.6K

Harry Coppock retweetledi

Xander Davies@alxndrdavies·7 Ağu

We at @AISecurityInst worked with @OpenAI to test GPT-5's safeguards. We identified multiple jailbreaks, including a universal jailbreak that evades all layers of mitigations and is being patched. Excited to continue partnering with OpenAI to test & strengthen safeguards.

English

128

22.8K

Harry Coppock retweetledi

Andy Zou@andyzou_jiaming·29 Tem

We deployed 44 AI agents and offered the internet $170K to attack them. 1.8M attempts, 62K breaches, including data leakage and financial loss. 🚨 Concerningly, the same exploits transfer to live production agents… (example: exfiltrating emails through calendar event) 🧵

English

387

2.2K

524.9K

Harry Coppock retweetledi

AI Security Institute@AISecurityInst·30 Tem

📢Introducing the Alignment Project: A new fund for research on urgent challenges in AI alignment and control, backed by over £15 million. ▶️ Up to £1 million per project ▶️ Compute access, venture capital investment, and expert support Learn more and apply ⬇️

English

195

123.5K

Harry Coppock retweetledi

Xander Davies@alxndrdavies·17 Tem

We at @AISecurityInst worked with @OpenAI to test & improve Agent’s safeguards prior to release. A few notes on our experience🧵 1/4

English

151

19.7K

Harry Coppock retweetledi

AI Security Institute@AISecurityInst·26 Haz

🧵 AI Systems are developing advanced cyber capabilities. This means they’re helping strengthen defences - but can also be used as threats. To keep on top of these risks, we need more rigorous evaluations of agentic AI, which is why we’re releasing Inspect Cyber 🔍

English

12.6K

Keşfet

@Irregular @meridianlabs_ai @soundboy @AISecurityInst @turinginst @AnthropicAI @AiEleuther @OpenAI