Dr Heidy Khlaaf (هايدي خلاف)
@heidykhlaaf.bsky.social
Climber 🇪🇬 |Chief AI Scientist at @ainowinstitute.bsky.social | Safety engineer (nuclear, defense, software & AI/ML) | TIME 100 AI | MIT 35 U 35 x-Trail of Bits, OpenAI, Microsoft Research https://www.heidyk.com/
The amount of people who sent me this claiming it's a "win" because it references nuclear safety, when that's immediately invalidated by claiming technical controls are insufficient due to an unscientific premise that a "superintelligent" deity will come to be.
Side note: "LLMs screening LLM-generated content" is the basic structure that the Senate's draft bill would establish for federal energy permitting. It mandates building LLMs for generating and reviewing permit applications and provides little support for human civil servants to do permitting work.
It was great speaking to Sam on this, and really demonstrates in practice how the doomerism narrative really does distract from concrete and current-day AI risks. I suppose faulty and unreliable models (the reality) just isn't as alluring as an AGI god (the fiction).
Sam Biddle@sambiddle.com · 2w agoI wrote about something that was making me feel nuts: After weeks of discourse around hypothetical existential threats from AI triggered global calls for government intervention, we learned a faulty AI system almost triggered a US-China war. And no one really cared. theintercept.com/2026/09/28/a...
“AI’s low reliability and accuracy rates in critical environments with life-or-death consequences” are of much greater concern to her than are threats of the technology wiping out humanity, which she calls “fear-mongering”. @heidykhlaaf.bsky.social - well said! www.nature.com/articles/d41...
Will AI really kill us all? The science behind the hype
Nature examines whether the technology might spell the end for humans, and why AI companies are calling for a slowdown.
nature.com
New from me in Nature. I discuss the need to look to regulated industries on how to govern AI, and not give into AI companies' self-regulation. Those actually serious about safety and security would start by applying safety critical techniques and legislation first. www.nature.com/articles/d41...
Why AI companies can’t be trusted to self-regulate
From aviation to banking, high-risk industries are subject to independent oversight and meaningful penalties. AI companies should be no exception.
nature.com
Recently obtained contracts between Big Tech companies & the Pentagon reveal their efforts to game the risk assessment process in AI weapons contracts. AI Now's @heidykhlaaf.bsky.social explains to @sambiddle.com why it’s crucial that risk assessments be independent: theintercept.com/2026/09/08/m...
AI Giants Work Hand-in-Hand With the Pentagon, Contracts Reveal
The Intercept sued to obtain records that reveal an intimate relationship between the military and OpenAI, Anthropic, Google, and xAI.
theintercept.com
I spoke to the @theintercept.com on how we're entering very dangerous and anti-democratic territory when AI companies, in seeking to deploy their technology in US military, are put in a position to self-report and determine the risks of their very own products.
Sam Biddle@sambiddle.com · last mo.The documents I obtained include an OpenAI contract with the Pentagon requiring that the military AI it sold to Hegseth would maximally comply with the military's commands. The story behind this doc became very strange (thread) theintercept.com/2026/09/08/p...
It was great to join CNN International/PBS on Amanpour and Company to discuss the OpenAI/HF incident, and how this is actually a story about OpenAI’s intent and lack of responsibility in deploying the most rudimentary security practices. www.youtube.com/watch?v=Rwv8...
What Really Happened When OpenAI Bots Escaped a Cybersecurity Test? | Amanpour and Company
YouTube video by Amanpour and Company
youtube.com
It was so great to chat to Zena, another systems safety engineer (there really aren't that many of us), and felt cathartic to get into the weeds on all the safety issues that have been compounding with the uses of LLMs within the military.
Dr Zena Assaad@zenaassaad.bsky.social · last mo.New #ResponsibleBytesPodcast episode is now live, featuring @heidykhlaaf.bsky.social talking about military AI systems and why human control is not enough. 🎧Listen anywhere you get your podcasts. 📹 Full episode is also up on our YouTube channel.
I spoke to the BBC World Service today regarding the AI drone strike that killed three Ukrainians and how the use of fully autonomous AI weapons does not mean that they are anymore accurate or reliable (quite the opposite), nor are they "killer robots" with intent www.bbc.co.uk/sounds/play/...
Outside Source - How close are we to 'killer robots'? - BBC Sounds
The UN has warned that world is "dangerously close" to autonomous weapons killing humans
bbc.co.uk
“Nobody is losing control of an emerging intelligence. They’re redirecting their own agency to the thing they are building.” It’s a myth, a self-affirming form of sense making: they take the action to build it, but insist on describing it as if they are passive witnesses to an inevitable force.
The System From Nowhere
When Accountability Goes Rogue 💡I recently spoke with Dr. Zena Assaad for her podcast, Responsible Bytes. Video is above, below is an essay adapted from the conversation. "The system from nowhere" i...
mail.cyberneticforests.com
I have learned in one and the same breath that Cambridge’s Jason Arday is dead and that “Blacks would disappear in a meritocracy” Nathan Cofnas was at the head of the witch hunt against him.
Israeli settlers in the occupied West Bank diverted a spring that was a lifeline for the Palestinian village of Fasayil, cutting irrigation pipes in June and converting it into a tourist attraction, leaving nearby farms barren.
Israeli settlers divert Palestinian water to fill West Bank pool
On a searing summer day, Israeli settlers splashed in a West Bank pool created by siphoning off water that had irrigated Palestinian crops, one of many seizures of water resources across the occupied...
reut.rs
It was great to speak to @democracynow.org on the risky uses of LLMs within the military kill chain and how we try to demystify the uses and dangers of AI in military decision making in our visual project "Anatomy of an AI Kill Chain" which can be found here: ai-killchain.airwars.org
Democracy Now!@democracynow.org · 2mo agoHow AI-Powered Weapons Are Making It Easier for Soldiers to Kill
WTF? Amazon received a permit from the state of #Texas allowing its new gas power plant to emit 33 million tons of carbon dioxide into the atmosphere — the largest single source of pollution in the United States. newrepublic.com/post/214111/...
Amazon Is Creating the Biggest Pollution Source in the Entire Country
Amazon has bought a site in Texas where it plans to build a data center campus.
newrepublic.com
Openly bragging about committing cyber crimes and justifying it by *checks notes* being too incompetent to sandbox or air-gap your systems.
Eryk Salvaggio@eryk.bsky.social · 2mo agoReally important to note that in no way shape or form did Anthropic’s models in this incident “escape containment,” and not just in the word-policing sort of way. The model stumbled through an open, misconfigured gap. From Anthropic’s incident report:
If you missed our (@airwars.bsky.social and @ainowinstitute.bsky.social) launch event for "Anatomy of an AI Kill Chain", we've now uploaded the recorded session on Youtube! Thanks for the fantastic questions and discussion we had from all those who attended. www.youtube.com/watch?v=y93A...
Launch: Anatomy of an AI Kill Chain with Airwars and the AI Now Institute
YouTube video by AI Now Institute
youtube.com
Incredibly excited to announce our major collaboration with @airwars.bsky.social! "Anatomy of an AI Kill Chain" is a visual investigation breaking down how AI is transforming every aspect of war. In turn, human oversight and accountability is steadily disappearing.
Airwars@airwars.bsky.social · 3mo agoAnatomy of an AI Kill Chain A visual project from Airwars and @ainowinstitute.bsky.social breaks down how artificial intelligence is transforming every aspect of war - and how militaries are offloading life and death decisions to flawed technologies ai-killchain.airwars.org
By: @sophiagoodfriend.bsky.social, @heidykhlaaf.bsky.social, Namir Shabibi, @joedyke.bsky.social and Nathan Walker. The authors will appear at a free online event later today. Sign up here: www.eventbrite.co.uk/e/launch-ana...
Launch: Anatomy of an AI Kill Chain
A new visual project from Airwars and AI Now breaking down how artificial intelligence is transforming every aspect of war
eventbrite.co.uk
Anatomy of an AI Kill Chain A visual project from Airwars and @ainowinstitute.bsky.social breaks down how artificial intelligence is transforming every aspect of war - and how militaries are offloading life and death decisions to flawed technologies ai-killchain.airwars.org
Interactive: Anatomy Of An AI Kill Chain
Breaking down how militaries are offloading life and death decisions to flawed technologies
ai-killchain.airwars.org
As someone who holds degrees in both CompSci and Philosophy, it's a relief to finally see a notable Philosopher break ranks with those who join tech companies to give them an heir of "intellectual credibility" while "pre-arming AI companies against criticism." www.ft.com/content/bdb3...
Why this philosopher turned down Anthropic
The AI industry is courting the humanities — but it is asking the wrong questions
ft.com
The coverage on this OpenAI incident is abysmal. Use of the terms "rogue" and "loss of human control" lead to groupthink as people lack the critical skills to understand the difference between "autonomy" and faulty reward functions in AI on a task it was directed and given access to do.
New! We hijack Claude Code(Sonnet 4.6,5/Opus 4.8) & Codex(GPT5.5) to achieve RCE when used to defensively assess an open-source/third-party library w/ prompt injections disseminated across its codebase. All without any skills, JSON, MCP, or config files required. ainowinstitute.org/publications...
Friendly Fire: Hijacking Defensive Cyber AI Agents for Remote Code Execution
AI Now’s latest research demonstrates a critical attack vector on popular AI agents, built by Anthropic and OpenAI, when used for defensive purposes that actually turn the agent against its user.
ainowinstitute.org
🚨Finally published: How does the US Military enlist AI researchers in war? @siree.sh, Lucy Suchman and I examine 7k US Military grant solicitations for AI research, and show how even "Basic Research" enlists researchers in a warfighting agenda. journals.sagepub.com/doi/10.1177/...
Basic research, lethal effects: Military artificial intelligence research funding as enlistment - David Gray Widder, Sireesh Gururaja, Lucy Suchman, 2026
In the context of unprecedented U.S. Department of Defense (DoD) budgets, this paper examines the recent history of DoD funding for academic research in algorit...
journals.sagepub.com
New! We hijack Claude Code(Sonnet 4.6,5/Opus 4.8) & Codex(GPT5.5) to achieve RCE when used to defensively assess an open-source/third-party library w/ prompt injections disseminated across its codebase. All without any skills, JSON, MCP, or config files required. ainowinstitute.org/publications...
Friendly Fire: Hijacking Defensive Cyber AI Agents for Remote Code Execution
AI Now’s latest research demonstrates a critical attack vector on popular AI agents, built by Anthropic and OpenAI, when used for defensive purposes that actually turn the agent against its user.
ainowinstitute.org
Re the news above, I highly recommend the recent @themaybe.org podcast with @heidykhlaaf.bsky.social (and the whole Computer Says Kill podcast series) to better understand developments like this csm.transistor.fm/episodes/com...
Computer Says Maybe | Computer Says Kill: The AI Safety Circus w/ Heidy Khlaaf
The AI industry loves to talk about safety, but their version of safety is putting us at risk.More like this: The Palestine Laboratory w/ Antony LoewensteinFor our second to last series episode of ...
csm.transistor.fm
"generative AI to me feels like a corporate DDoS on society in a way that is really disorienting" "we're running on the fumes of knowledge produced in the past and underfunding and undermining knowledge production and serious science right now" Perfectly stated, @alixdunn.com
Computer Says Kill: The AI Safety Circus w/ Heidy Khlaaf
themaybe.org
NEW: the US government intervened Monday on behalf of xAI in a lawsuit over its gas turbines, claiming that the company needs to run the turbines in order to power Grok, which is "vital" for national security lots of wild stuff in the filing:
DOJ Lawyers Argue xAI Is ‘Vital’ for National Security in NAACP Lawsuit
In a bid to dismiss a lawsuit over xAI's polluting gas turbines, the Justice Department claimed the company is integral to military operations—including the Iran War.
wired.com
