Dr Heidy Khlaaf (هايدي خلاف)
@heidykhlaaf.bsky.social
Climber 🇪🇬 |Chief AI Scientist at @ainowinstitute.bsky.social | Safety engineer (nuclear, software & AI/ML) | TIME 100 AI
|x- Trail of Bits, Zipline, OpenAI, Adelard, MSFTResearch
https://www.heidyk.com/
|x- Trail of Bits, Zipline, OpenAI, Adelard, MSFTResearch
https://www.heidyk.com/
Pinned
New paper with @smw.bsky.social & @meredithmeredith.bsky.social. We challenge the narrative emphasising AI bioweapons risks, and bring attention to the covert proliferation of military intelligence, surveillance, targeting, and reconnaissance (ISTAR) already occurring via foundation models. 1/5
Despite warnings in our report, today's release of the UK Nuclear Regulatory Review is littered with unsubstantiated claims and recommendations touting AI "as a powerful tool" and "cost-effective" to be used for safety and licensing without noted risks or caveats. This trend has now reached the UK.
Problem: AI needs massive amounts of power to thrive. Nuclear makes lots of power. Nuclear takes a long long time to do safely.
Proposed solution that I'm sure will have no unpleasant consequences: Use AI to speed up the construction of new nuclear plants.
Proposed solution that I'm sure will have no unpleasant consequences: Use AI to speed up the construction of new nuclear plants.
Power Companies Are Using AI To Build Nuclear Power Plants
Tech companies are betting big on nuclear energy to meet AIs massive power demands and they're using that AI to speed up the construction of new nuclear power plants.
www.404media.co
November 24, 2025 at 1:16 PM
Despite warnings in our report, today's release of the UK Nuclear Regulatory Review is littered with unsubstantiated claims and recommendations touting AI "as a powerful tool" and "cost-effective" to be used for safety and licensing without noted risks or caveats. This trend has now reached the UK.
Reposted by Dr Heidy Khlaaf (هايدي خلاف)
I've said it before and I will say it again. There is no way to secure a system when it's potential attack surface is *all of language*
Looks like LLMs are *very* vulnerable to attack via poetic allusion: "curated poetic prompts yielded high attack-success rates (ASR), with some providers exceeding 90% ..."
https://arxiv.org/html/2511.15304v1
https://arxiv.org/html/2511.15304v1
November 20, 2025 at 5:23 PM
I've said it before and I will say it again. There is no way to secure a system when it's potential attack surface is *all of language*
Reposted by Dr Heidy Khlaaf (هايدي خلاف)
Tech companies are betting big on nuclear energy to meet AI’s massive power demands—and Trump’s done a lot to make it easier for them. Heidy Khlaaf, the head AI scientist at the AI Now Institute, tells us why that’s dangerous.
@mjgault.bsky.social has the story:
www.404media.co/power-compan...
@mjgault.bsky.social has the story:
www.404media.co/power-compan...
November 14, 2025 at 6:57 PM
Tech companies are betting big on nuclear energy to meet AI’s massive power demands—and Trump’s done a lot to make it easier for them. Heidy Khlaaf, the head AI scientist at the AI Now Institute, tells us why that’s dangerous.
@mjgault.bsky.social has the story:
www.404media.co/power-compan...
@mjgault.bsky.social has the story:
www.404media.co/power-compan...
Great coverage by @mjgault.bsky.social on our report and what's at stake and what could go wrong in using AI in an attempt to accelerate nuclear development. Read our report here: ainowinstitute.org/publications...
NEW: Tech companies are betting big on nuclear energy to meet AIs massive power demands and they're using that AI to speed up the construction of new nuclear power plants.
And despite expert concerns about potential disaster, the US government is on board.
And despite expert concerns about potential disaster, the US government is on board.
Power Companies Are Using AI To Build Nuclear Power Plants
Tech companies are betting big on nuclear energy to meet AIs massive power demands and they're using that AI to speed up the construction of new nuclear power plants.
www.404media.co
November 14, 2025 at 6:52 PM
Great coverage by @mjgault.bsky.social on our report and what's at stake and what could go wrong in using AI in an attempt to accelerate nuclear development. Read our report here: ainowinstitute.org/publications...
Reposted by Dr Heidy Khlaaf (هايدي خلاف)
This fast-tracking approach comes alongside efforts from many of these AI companies themselves to apply unproven AI systems to speed the pace of licensing/regulation. It also forms the core of a new report from the @ainowinstitute.bsky.social @heidykhlaaf.bsky.social
November 14, 2025 at 3:51 PM
This fast-tracking approach comes alongside efforts from many of these AI companies themselves to apply unproven AI systems to speed the pace of licensing/regulation. It also forms the core of a new report from the @ainowinstitute.bsky.social @heidykhlaaf.bsky.social
Reposted by Dr Heidy Khlaaf (هايدي خلاف)
New: I wrote about the nuclear push coming from an energy-constrained AI industry, something that’s importantly coupled with an increasingly de-regulatory environment — often mirroring the language coming directly from these corporations — coming from the White House.
puck.news/ais-nuclear-...
puck.news/ais-nuclear-...
A.I. Goes Nuclear!
OpenAI, Google, and Microsoft are betting big on nuclear energy to power their A.I. data centers. But weakened regulations may create risks, nuclear safety experts warn.
puck.news
November 14, 2025 at 3:51 PM
New: I wrote about the nuclear push coming from an energy-constrained AI industry, something that’s importantly coupled with an increasingly de-regulatory environment — often mirroring the language coming directly from these corporations — coming from the White House.
puck.news/ais-nuclear-...
puck.news/ais-nuclear-...
Reposted by Dr Heidy Khlaaf (هايدي خلاف)
This week Open AI walked back a call for the govt to backstop financing for its trillion dollar investments in data centers. This was only the tip of the iceberg; a slow bailout for AI firms is already underway. Read more from @ambakak.bsky.social and I in @wsj.com: www.wsj.com/opinion/you-...
Opinion | You May Already Be Bailing Out the AI Business
Washington is treating the industry as if it’s too big to fail, even as the market sends lukewarm signals.
www.wsj.com
November 12, 2025 at 10:56 PM
This week Open AI walked back a call for the govt to backstop financing for its trillion dollar investments in data centers. This was only the tip of the iceberg; a slow bailout for AI firms is already underway. Read more from @ambakak.bsky.social and I in @wsj.com: www.wsj.com/opinion/you-...
New Report: Fission for Algorithms. We draw on our nuclear expertise to dissect the risky fast-tracking initiatives hastening nuclear development in service of AI. This includes proposals to use Gen AI for nuclear licensing, whilst lowering well-established nuclear thresholds.
Fission for Algorithms: The Undermining of Nuclear Regulation in Service of AI - AI Now Institute
A report examining nuclear “fast-tracking” initiatives on their feasibility and their impact on nuclear safety, security, and safeguards.
ainowinstitute.org
November 12, 2025 at 11:05 AM
New Report: Fission for Algorithms. We draw on our nuclear expertise to dissect the risky fast-tracking initiatives hastening nuclear development in service of AI. This includes proposals to use Gen AI for nuclear licensing, whilst lowering well-established nuclear thresholds.
"Rafael purchased AI technologies made available through AWS, including the state-of-the-art large language model Claude ... The materials reviewed also indicate Amazon sold cloud-computing services to Israel’s nuclear program and offices administering the West Bank"
SCOOP: I obtained internal documents showing Amazon has been selling cloud computing and AI services to the state-owned Israeli weapons manufacturers whose bombs and missiles have ravaged Gaza.
theintercept.com/2025/10/24/a...
theintercept.com/2025/10/24/a...
As Israel Bombed Gaza, Amazon Did Business With Its Bomb-Makers
The Intercept has learned that Amazon sold cloud services to Israeli weapons firms at the height of Israel’s bombardment of Gaza.
theintercept.com
October 25, 2025 at 8:24 PM
"Rafael purchased AI technologies made available through AWS, including the state-of-the-art large language model Claude ... The materials reviewed also indicate Amazon sold cloud-computing services to Israel’s nuclear program and offices administering the West Bank"
Reposted by Dr Heidy Khlaaf (هايدي خلاف)
Anthropic’s partnership with the DOE to keep Claude from building a nuclear weapon makes for good headlines. @heidykhlaaf.bsky.social calls it security theater. The real risk is AI firms gaining access to national security data.
www.wired.com/story/anthro...
www.wired.com/story/anthro...
Anthropic Has a Plan to Keep Its AI From Building a Nuclear Weapon. Will It Work?
Anthropic partnered with the US government to create a filter meant to block Claude from helping someone build a nuke. Experts are divided on whether its a necessary protection—or a protection at all.
www.wired.com
October 22, 2025 at 5:48 PM
Anthropic’s partnership with the DOE to keep Claude from building a nuclear weapon makes for good headlines. @heidykhlaaf.bsky.social calls it security theater. The real risk is AI firms gaining access to national security data.
www.wired.com/story/anthro...
www.wired.com/story/anthro...
I spoke to @mjgault.bsky.social in WIRED on what I ultimately view as safety theatre for "nuclear safeguarding" and how it distracts from the real risk of unregulated private corporations having access incredibly sensitive nuclear secrets given their insecure AI models.
Anthropic says its AI won't help you build a nuclear weapon. Will it work? And can a chatbot even help build a nuke?
Anthropic Has a Plan to Keep Its AI From Building a Nuclear Weapon. Will It Work?
Anthropic partnered with the US government to create a filter meant to block Claude from helping someone build a nuke. Experts are divided on whether its a necessary protection—or a protection at all.
www.wired.com
October 21, 2025 at 9:58 AM
I spoke to @mjgault.bsky.social in WIRED on what I ultimately view as safety theatre for "nuclear safeguarding" and how it distracts from the real risk of unregulated private corporations having access incredibly sensitive nuclear secrets given their insecure AI models.
Reposted by Dr Heidy Khlaaf (هايدي خلاف)
Anthropic says its AI won't help you build a nuclear weapon. Will it work? And can a chatbot even help build a nuke?
Anthropic Has a Plan to Keep Its AI From Building a Nuclear Weapon. Will It Work?
Anthropic partnered with the US government to create a filter meant to block Claude from helping someone build a nuke. Experts are divided on whether its a necessary protection—or a protection at all.
www.wired.com
October 20, 2025 at 2:04 PM
Anthropic says its AI won't help you build a nuclear weapon. Will it work? And can a chatbot even help build a nuke?
Reposted by Dr Heidy Khlaaf (هايدي خلاف)
OpenAI, Anthropic & others have shifted from championing ethics to signing $200M+ defense contracts that embed gen AI into high-risk military systems. In @theverge.com, @heidykhlaaf.bsky.social explains why the move toward defense partnerships is a safety risk.
Listen here: shorturl.at/mTAZq
Listen here: shorturl.at/mTAZq
September 30, 2025 at 6:19 PM
OpenAI, Anthropic & others have shifted from championing ethics to signing $200M+ defense contracts that embed gen AI into high-risk military systems. In @theverge.com, @heidykhlaaf.bsky.social explains why the move toward defense partnerships is a safety risk.
Listen here: shorturl.at/mTAZq
Listen here: shorturl.at/mTAZq
I'm sorry what? We're now back to British colonial officers in British Mandate Palestine who hold “supreme political and legal authority”?
Tony Blair proposed as chief of Gaza transitional authority, reports say - https://www.middleeasteye.net/news/tony-blair-proposed-chief-gaza-transitional-authority-reports-say
Tony Blair proposed as chief of Gaza transitional authority, reports say
BBC reports that former UK prime minister has held talks about leading the interim authority
www.middleeasteye.net
September 26, 2025 at 8:57 AM
I'm sorry what? We're now back to British colonial officers in British Mandate Palestine who hold “supreme political and legal authority”?
It was great speaking to @haydenfield.bsky.social on the Decoder podcast, where we discussed the relationship between safety and the push for generative AI to be deployed within military applications.
www.theverge.com/podcast/7848...
www.theverge.com/podcast/7848...
How AI safety took a backseat to military money
Heidy Khlaff discusses the industry shift toward military applications and what it means for AI safety.
www.theverge.com
September 25, 2025 at 4:38 PM
It was great speaking to @haydenfield.bsky.social on the Decoder podcast, where we discussed the relationship between safety and the push for generative AI to be deployed within military applications.
www.theverge.com/podcast/7848...
www.theverge.com/podcast/7848...
Reposted by Dr Heidy Khlaaf (هايدي خلاف)
"The State of Israel has committed genocide."
Navi Pillay, chair of the UN Independent International Commission of Inquiry on the Occupied Palestinian Territory, has told Al Jazeera that Israel’s war on Gaza is a genocide.
Navi Pillay, chair of the UN Independent International Commission of Inquiry on the Occupied Palestinian Territory, has told Al Jazeera that Israel’s war on Gaza is a genocide.
September 16, 2025 at 7:57 AM
"The State of Israel has committed genocide."
Navi Pillay, chair of the UN Independent International Commission of Inquiry on the Occupied Palestinian Territory, has told Al Jazeera that Israel’s war on Gaza is a genocide.
Navi Pillay, chair of the UN Independent International Commission of Inquiry on the Occupied Palestinian Territory, has told Al Jazeera that Israel’s war on Gaza is a genocide.
Given that the US passed multiple EO to lower the thresholds of nuclear safety, it's concerning that "Britain and the US will accept each other’s safety checks on reactor designs, almost halving the time for a nuclear project to be licensed to about two years."
www.theguardian.com/business/202...
www.theguardian.com/business/202...
UK and US line up string of deals to build modular nuclear reactors in Britain
Agreements include plan to build 12 reactors in Hartlepool with Centrica, creating 2,500 jobs, and fast-tracking UK and US safety checks
www.theguardian.com
September 15, 2025 at 1:35 PM
Given that the US passed multiple EO to lower the thresholds of nuclear safety, it's concerning that "Britain and the US will accept each other’s safety checks on reactor designs, almost halving the time for a nuclear project to be licensed to about two years."
www.theguardian.com/business/202...
www.theguardian.com/business/202...
Being recognised in MIT's 35 Under 35 just a week after being in TIME100 AI is such an honour! The profile gets to the heart of the motivation of my work, which includes the use of AI in Gaza that has contributed to a devastating death toll: www.technologyreview.com/innovator/he...
September 8, 2025 at 12:49 PM
Being recognised in MIT's 35 Under 35 just a week after being in TIME100 AI is such an honour! The profile gets to the heart of the motivation of my work, which includes the use of AI in Gaza that has contributed to a devastating death toll: www.technologyreview.com/innovator/he...
Reposted by Dr Heidy Khlaaf (هايدي خلاف)
We are excited to announce that AI Now’s Chief AI Scientist, Dr. @heidykhlaaf.bsky.social, is one of MIT @technologyreview.com's 2025 Innovators Under 35, which recognizes scientists, inventors, and entrepreneurs from around the world.
Read about Dr. Khlaaf’s work here: ter.li/5iqt4y
Read about Dr. Khlaaf’s work here: ter.li/5iqt4y
September 8, 2025 at 12:42 PM
We are excited to announce that AI Now’s Chief AI Scientist, Dr. @heidykhlaaf.bsky.social, is one of MIT @technologyreview.com's 2025 Innovators Under 35, which recognizes scientists, inventors, and entrepreneurs from around the world.
Read about Dr. Khlaaf’s work here: ter.li/5iqt4y
Read about Dr. Khlaaf’s work here: ter.li/5iqt4y
Reposted by Dr Heidy Khlaaf (هايدي خلاف)
We are excited to announce that AI Now’s Chief AI Scientist, @heidykhlaaf.bsky.social, is included on this year’s #TIME100AI list, which features leaders, policymakers, artists, and entrepreneurs who are advancing major conversations about how AI is reshaping the world: time.com/collections/...
TIME100 AI 2025: Heidy Khlaaf
Find out why Heidy Khlaaf made TIME’s list of the most influential people in artificial intelligence
time.com
August 28, 2025 at 1:36 PM
We are excited to announce that AI Now’s Chief AI Scientist, @heidykhlaaf.bsky.social, is included on this year’s #TIME100AI list, which features leaders, policymakers, artists, and entrepreneurs who are advancing major conversations about how AI is reshaping the world: time.com/collections/...
Excited to be included on TIME 's list of the #TIME100 AI of 2025!
Grateful to @ainowinstitute.bsky.social for supporting my work on addressing the safety revisionism of AI in defense/safety-critical contexts where safety has been obscured with little to no accountability.
time.com/collections/...
Grateful to @ainowinstitute.bsky.social for supporting my work on addressing the safety revisionism of AI in defense/safety-critical contexts where safety has been obscured with little to no accountability.
time.com/collections/...
TIME100 AI 2025: Heidy Khlaaf
Find out why Heidy Khlaaf made TIME’s list of the most influential people in artificial intelligence
time.com
August 28, 2025 at 12:12 PM
Excited to be included on TIME 's list of the #TIME100 AI of 2025!
Grateful to @ainowinstitute.bsky.social for supporting my work on addressing the safety revisionism of AI in defense/safety-critical contexts where safety has been obscured with little to no accountability.
time.com/collections/...
Grateful to @ainowinstitute.bsky.social for supporting my work on addressing the safety revisionism of AI in defense/safety-critical contexts where safety has been obscured with little to no accountability.
time.com/collections/...
Reposted by Dr Heidy Khlaaf (هايدي خلاف)
As the US considers acquiring agentic AI systems to produce AI-generated propaganda campaigns, @heidykhlaaf.bsky.social warned @sambiddle.com of @theintercept.com that these systems are easily compromised, threatening national security rather than protecting it:
theintercept.com/2025/08/25/p...
theintercept.com/2025/08/25/p...
Pentagon Document: U.S. Wants to “Suppress Dissenting Arguments” Using AI Propaganda
The U.S. is interested in acquiring machine-learning technology to carry out AI-generated propaganda campaigns overseas.
theintercept.com
August 26, 2025 at 1:50 PM
As the US considers acquiring agentic AI systems to produce AI-generated propaganda campaigns, @heidykhlaaf.bsky.social warned @sambiddle.com of @theintercept.com that these systems are easily compromised, threatening national security rather than protecting it:
theintercept.com/2025/08/25/p...
theintercept.com/2025/08/25/p...
I spoke to the Intercept on the use of agentic AI for disinformation campaigns, and how SOCOM framing it as a mitigation to adversaries' "use" is a misrepresentation, as offensive and defensive uses are two sides of the same coin that allows them to use it in the same way.
U.S. Special Operations Command, recently implicated in a covert anti-vax campaign targeting Asian countries, is seeking machine learning tools to automate digital propaganda and “suppress dissenting arguments" online, according to a procurement document I obtained theintercept.com/2025/08/25/p...
Pentagon Document: U.S. Wants to “Suppress Dissenting Arguments” Using AI Propaganda
The U.S. is interested in acquiring machine-learning technology to carry out AI-generated propaganda campaigns overseas.
theintercept.com
August 26, 2025 at 9:31 AM
I spoke to the Intercept on the use of agentic AI for disinformation campaigns, and how SOCOM framing it as a mitigation to adversaries' "use" is a misrepresentation, as offensive and defensive uses are two sides of the same coin that allows them to use it in the same way.
Reposted by Dr Heidy Khlaaf (هايدي خلاف)
Hind Khoudary explains what happened in the attack on Nasser hospital.
An explosive drone targeted two civilians, one of whom was a journalist. Then, as rescuers went to retrieve the bodies and additional journalists were filming what had happened, a missile targeted them.
An explosive drone targeted two civilians, one of whom was a journalist. Then, as rescuers went to retrieve the bodies and additional journalists were filming what had happened, a missile targeted them.
August 25, 2025 at 8:20 AM
Hind Khoudary explains what happened in the attack on Nasser hospital.
An explosive drone targeted two civilians, one of whom was a journalist. Then, as rescuers went to retrieve the bodies and additional journalists were filming what had happened, a missile targeted them.
An explosive drone targeted two civilians, one of whom was a journalist. Then, as rescuers went to retrieve the bodies and additional journalists were filming what had happened, a missile targeted them.
Reposted by Dr Heidy Khlaaf (هايدي خلاف)
Fuck the IOF forever.
After bombing the hospital, their airstrike murders the medics and journalists who arrive to assist/document.
No words...
After bombing the hospital, their airstrike murders the medics and journalists who arrive to assist/document.
No words...
August 25, 2025 at 9:47 AM
Fuck the IOF forever.
After bombing the hospital, their airstrike murders the medics and journalists who arrive to assist/document.
No words...
After bombing the hospital, their airstrike murders the medics and journalists who arrive to assist/document.
No words...