Our New Age of Artificial, Intelligent Surveillance
Added 2025-07-12 12:06:10 +0000 UTCNo one is talking about how LLMs enable a new era of mass intelligent surveillance. This is not just Grok 4 @ Doge, this is world-wide, as we speak. The second documentary, exclusive, early and ad-free on Patreon.
Gemini (10m tokens): https://arxiv.org/abs/2403.05530
NYT: Anthropic CEO on regulation and transparency: https://www.nytimes.com/2025/06/05/opinion/anthropic-ceo-regulate-transparency.html
Paper on in-context scheming reasoning: https://static1.squarespace.com/static/6593e7097565990e65c886fd/t/6751eb240ed3821a0161b45b/1733421863119/in_context_scheming_reasoning_paper.pdf
Anthropic Claude 4 ethics and behavior PDF: https://www-cdn.anthropic.com/6be99a52cb68eb70eb9572b4cafad13df32ed995.pdf
Brydon Eastman X post on AI moral judgements: https://x.com/brhydon/status/1926077805185429973
Sam Altman video clip on X: https://x.com/vitrupo/status/1930009915650912586
Zuckerberg wants AI that "knows us well": https://www.thebignewsletter.com/p/monopoly-round-up-mark-zuckerberg
NYT: Trump, Palantir, and data on Americans: https://www.nytimes.com/2025/05/30/technology/trump-palantir-data-americans.html
YouTube video on AI drug discovery: https://www.youtube.com/watch?v=-KFO0pES-zQ
Isomorphic Labs website: https://www.isomorphiclabs.com/
Axios: State Department AI reviews student visas: https://www.axios.com/2025/03/06/state-department-ai-revoke-foreign-student-visas-hamas
New Scientist: AI hallucinations are getting worse: https://www.newscientist.com/article/2479545-ai-hallucinations-are-getting-worse-and-theyre-here-to-stay/
Axios: State Department official on AI tools: https://www.axios.com/2025/03/06/state-department-ai-revoke-foreign-student-visas-hamas
Reuters: Musk's DOGE using AI to snoop: https://www.reuters.com/technology/artificial-intelligence/musks-doge-using-ai-snoop-us-federal-workers-sources-say-2025-04-08/
Reuters: DOGE utilizing Grok models for surveillance: https://www.reuters.com/technology/artificial-intelligence/musks-doge-using-ai-snoop-us-federal-workers-sources-say-2025-04-08/
NYT: DOGE spending cuts and AI hallucinations: https://www.nytimes.com/2025/02/18/upshot/doge-contracts-musk-trump.html?unlocked_article_code=1.yE4.WGHU.oOR4mey5T-XE&smid=url-share
NBC News: Federal workers push back on Musk email: https://www.nbcnews.com/politics/doge/federal-workers-agencies-push-back-elon-musks-email-ultimatum-rcna193439
Reuters: Concerns over Musk using AI gathered information: https://www.reuters.com/technology/artificial-intelligence/musks-doge-using-ai-snoop-us-federal-workers-sources-say-2025-04-08/
MIT Tech Review: Worker monitoring platforms: https://www.technologyreview.com/2025/02/24/1111664/worker-monitoring-employee-surveillance/#:~:text=%E2%80%9CWe're%20at%20a%20crucial,do%20it%2C%E2%80%9D%20she%20says.
Computerworld: AI bossware predicting employee behaviors: https://www.computerworld.com/article/3843213/how-ai-enabled-bossware-is-being-used-to-track-and-evaluate-your-work.html
MI6 and CIA chiefs using generative AI: https://www.publictechnology.net/2024/09/10/defence-and-security/mi6-and-cia-chiefs-we-are-using-generative-ai-to-identify-key-information-in-a-sea-of-data/
Liberty Human Rights on UK mass surveillance: https://www.libertyhumanrights.org.uk/fundamental/mass-surveillance-snoopers-charter/
FT: UK is "wild west" for live face scanning: https://www.ft.com/content/09ce1755-554e-43c9-90ad-8e07196763b0
Fox News: AI enabling Iran's crackdown on women: https://www.foxnews.com/world/ai-enabling-irans-crackdown-on-women-authoritarian-regime-uses-tech-enforce-head-covering
The Guardian: Israel military AI surveillance in West Bank: https://www.theguardian.com/world/2025/mar/06/israel-military-ai-surveillance#:~:text=Israel%E2%80%99s%20military%20surveillance%20agency%20has,by%20the%20Guardian%20can%20reveal
The Guardian: Israeli model required "psychotic amounts" of data: https://www.theguardian.com/world/2025/mar/06/israel-military-ai-surveillance#:~:text=Israel%E2%80%99s%20military%20surveillance%20agency%20has,by%20the%20Guardian%20can%20reveal
The Guardian: AI makes blanket surveillance more effective: https://www.theguardian.com/world/2025/mar/06/israel-military-ai-surveillance#:~:text=Israel%E2%80%99s%20military%20surveillance%20agency%20has,by%20the%20Guardian%20can%20reveal
The Guardian: AI identifying words indicating "troublemaking": https://www.theguardian.com/world/2025/mar/06/israel-military-ai-surveillance#:~:text=Israel%E2%80%99s%20military%20surveillance%20agency%20has,by%20the%20Guardian%20can%20reveal
The Guardian: AI allows control over the population: https://www.theguardian.com/world/2025/mar/06/israel-military-ai-surveillance#:~:text=Israel%E2%80%99s%20military%20surveillance%20agency%20has,by%20the%20Guardian%20can%20reveal
PRC LLM content examples dataset: https://www.documentcloud.org/documents/25599140-prc-llm-content-examples-4/
TechCrunch: Chinese government improving repression with LLMs: https://techcrunch.com/2025/03/26/leaked-data-exposes-a-chinese-ai-censorship-machine/
Substack post on leaked Chinese database: https://substack.com/home/post/p-155226006
TechCrunch: China AI flagging sensitive political topics: https://techcrunch.com/2025/03/26/leaked-data-exposes-a-chinese-ai-censorship-machine/
TechCrunch: LLMs improving efficiency of information control: https://techcrunch.com/2025/03/26/leaked-data-exposes-a-chinese-ai-censorship-machine/
RFA: China surveillance recognizing Uighur ethnicity: https://www.rfa.org/english/news/china/profiling-12222020103743.html
SCMP: China AI analyzing shopping records, search histories: https://www.scmp.com/news/china/politics/article/3310749/chinas-next-gen-surveillance-tools-get-ai-boost-target-telegram-and-vpn-users
EU AI Act Article 5: Facial recognition databases: https://artificialintelligenceact.eu/article/5/
EU AI Act Article 5: Emotion recognition prohibitions: https://artificialintelligenceact.eu/article/5/
CCN: Sam Altman on AI safety and mass surveillance: https://www.ccn.com/news/technology/sam-altman-ai-safety-mass-surveillance/
Dario Amodei essay on AI propaganda and surveillance: https://www.darioamodei.com/essay/machines-of-loving-grace
The Verge: Anthropic working on Claude Gov model: https://www.theverge.com/ai-artificial-intelligence/680465/anthropic-claude-gov-us-government-military-ai-model-launch
NYT: OpenAI working with UAE data centers: https://www.nytimes.com/2025/05/22/technology/openai-uae-data-centers.html
Comments
Let’s publish this on YouTube? Such an important topic!
Petr Sektonov
2025-07-26 07:38:07 +0000 UTCExcellent and interesting analysis, as always! One nitpick, however: on your maps, you've not labeled Iran. That's Afghanistan.
Ethan Haluza-Delay
2025-07-21 23:32:27 +0000 UTCMaybe I missed something in the video...where did Philip say anything that contradicts your main point that dictatorships are, "indications of a societal breakdown across all its layers" ? Are you saying the available tools/technology available to a dictatorship in their current time have no effect on the way they consolidate power and suppress dissent? That available tools and technology have no bearing at all on the degree or nature of state suppression or surveillance? "My conclusion is that surveillance state will arrive not when a super-evil president equipped with a super-AI grabs the power" I don't think the video suggested otherwise. How could it suggest that anyway when we have literal current-day examples of several surveillance states that have been existing well before this current era of Generative AI? The video says nothing about "arrival" or ontologies of dictatorships and is just talking about how AI is being used for arguably dictatorial means (not "ends", but means). I think the video is very clearly just gathering several actual sources together to highlight how Generative AI is currently being deployed with the purpose to surveil the masses and how the intent behind that surveillance is clearly for the sake of suppressing dissent and/or consolidating state power.
hyzypg
2025-07-21 17:28:18 +0000 UTCIn the words of Ilya Sutskever: "AI could enable perpetually stable dictatorships.".
Erik
2025-07-18 22:03:19 +0000 UTCI must admit that I disagree with the main thesis of this video (but thanks for the great research anyway, Philip) and the majority of other comments. Dictatorship and surveillance states are not functions of technology or an evil person on top, but are indications of a societal breakdown across all its layers. I realized this while studying Soviet repressions in 1920s and 1930s. Although Stalin as the country leader definitely bears the main responsibility, it was evident that many people, including those at the bottom of the society, found the repressions useful. There were three main groups of those: (1) Marxist idealists who saw all the non-conformists as a threat to the bright communistic future; (2) those who were solving personal problems (reporting the boss to get promoted, reporting the husband of the lover, etc.); (3) those who pursued safe and prestigious careers in the police, army, government etc. So, we speak about a broad meltdown of the society, not just one evil person. And in my subjective opinion, it was much more horrible to be snitched by somebody who pretended to be your friend than by some AI that listens 24/7 through your phone. Note that the Soviet repressions abaded with time from their 1937-1938 peak (even during the remaining Stalin rule) despite the emergence of new surveillance technologies. My conclusion is that surveillance state will arrive not when a super-evil president equipped with a super-AI grabs the power, but when broad swaths of the society find such a state to be beneficial for them personally. My recommendation: (1) don't be lured by the promises of better life if we only eliminate XYZ people, whoever XYZ might be, and (2) don't resort to harming other people to obtain personal benefits and build the career.
Roman
2025-07-16 18:17:01 +0000 UTCYes, German secret services, for example, were profiling EU residents and dodging the GDPR in other ways. So, I assume that the AI Act is not an obstacle either. And given that foreign spies, hackers etc. also most probably use LLMs, I don't understand how the EU intelligence agencies can go without.
Roman
2025-07-16 17:35:56 +0000 UTCNot good enough to do that much good. But good enough to do a lot of harm. There’s no contradiction and it is not a laughing matter, Youssef!
Pavol Vaskovic
2025-07-16 08:16:08 +0000 UTCCan you be more specific about the backports and efforts to ban end-to-end encryption? I wasn’t aware and would like to research more. Thanks!
Pavol Vaskovic
2025-07-16 07:51:08 +0000 UTCI was going to say this is one of the most important videos on Patreon - but it's actually one of the most important videos online today. And I say that realizing it sounds hyperbolic. People need this information. Everyone does. The other problem, though, is that everyone is at the same time perpetually distracted by design thanks to the attention economy. We are truly in remarkable times. Thank you for doing this, I appreciate the work and the willingness to put this out there.
Mike Hindes
2025-07-16 04:05:03 +0000 UTCHead, meet sand.
Mike Hindes
2025-07-16 04:01:51 +0000 UTCYou're a complete doomer. I'm considering rescinding my subscription at this point. It has been nothing but negativity from Phillip for YEARS. Zero mention of what immense good AI could do for society for YEARS. I'm just sick of the incessant negativity that seems to have consumed the AI commentary space.
r
2025-07-15 22:24:11 +0000 UTCWell if history can teach us anything the mass disobedience is the only way. At the end of comunist era there was not need to find out who is against government - everyone was to some degree - but there was no way to arrest everyone. Saying that I'm a bit pesymistic. If in USA an idiot can persuade many people to oppress others what AI will be able to do.
Arek Stryjski
2025-07-15 17:14:16 +0000 UTCThank you for making this video! ❤️ Again fantastic quality and commentary. I intend to use it as a conversation starter in my social circles. [Edit: Though I’m sad to see that there is no link to share the video with. Limits who I can show this too. 😢] I’ve been worried about this kind of threat for more than a year now and am super concerned that if AGI doesn’t end up killing us and can be controlled, the most probable scenario is a mass-surveillance dystopia that’s God-knows-how-stable and also unprecedented in its ability to ignore the needs of its citizens. I fear that this could happen regardless of where you live in the world as the digital spheres are controlled by China and the US and so they will have the power to extend their grasp on other countries. The prospect of a permanent global AI-fueled mass-surveillance dystopia is currently even scarier to me than direct extinction of humanity due to AI and I wonder what we could best do to prevent that.
Birk Källberg
2025-07-15 07:13:00 +0000 UTCid love to get your input with the guys over at Big Brother Watch, they are doing amazing work but your technical expertise would only be a good thing for everyone
Nick
2025-07-14 17:40:07 +0000 UTCScary stuff Phillip. Another great video. I believe there is a Chinese "curse" ... "may you live in interesting times"... well...
Daniel A Barbatti
2025-07-14 14:56:19 +0000 UTCFantastic video on this paradigm shift in surveillance. Thanks for sharing your thoughts on it. However, as someone who values their privacy quite a lot your call to action against self-censorship has me quite conflicted. I just don't quite believe in millions of "needles" sticking together to counteract this, once the loudest voices starts being targeted and silenced the others will fall in line. However, I strongly believe that raising awareness about this and sharing your message is incredibly important so I really hope you'll be posting this one on Youtube as well!
Håvard Rekdal
2025-07-14 10:23:18 +0000 UTCThought people might enjoy this related programme from back in the day. https://www.youtube.com/watch?v=jRTv9S8ufBw 1980: Will TECHNOLOGY put an END to JOBS? | General Studies | Predicting the Future | BBC Archive "The capacity to control society because of this technology is almost limitless, unfortunately - that's why we have to be so vigilant it's untrue. How will the silicon chip affect the quality of our lives and our attitudes to work? Starting at the hyper-modern Washington Metro in the US, Bernard Falk visits places that have already adopted new technology to find out what could be in store for white collar, blue collar and professional workers in the next decade. What kinds of jobs will change or disappear? Are there areas where employment will increase? What will be the effects on various professions? How must our attitudes change if we are to enjoy the benefits of this new technological revolution and not suffer its worst effects?"
Billy
2025-07-13 15:41:58 +0000 UTCSo since the hype is cooling down Fear mongering might seem like a good idea Oh no it is not that the AI systems aren’t good enough It is that they could destroy the world so we decided to give up on them Lol
Youssef Mohamed
2025-07-13 10:59:42 +0000 UTCGreat video once again. Very thought provoking. Thanks for making me aware of this.
clay-loop
2025-07-12 23:08:45 +0000 UTCI wonder if AI can be used as a tool to subvert this kind of surveillance. Perhaps there will end up being an AI arms race between those being monitored and those doing the monitoring.
Barnaby Golden
2025-07-12 21:31:33 +0000 UTCThank you (again and continuously :) for putting together clear headed, intelligent takes on the state of AI development. I appreciate that it's not just the technical implementation, but also societal impact.
Alexi Kostibas
2025-07-12 18:19:46 +0000 UTC1984
Youssef Mohamed
2025-07-12 16:40:13 +0000 UTCExcellent video. I believe the EU AI act regulations contain opaquely worded "carve outs" for anything related to intelligence or national security. This includes INTERPOL or non-EU police forces executing mutual-legal-assistance or evidence-sharing treaties. It says the regulations "should not apply to public authorities of third countries … when acting in the framework of cooperation or international agreements for law-enforcement and judicial cooperation with the Union or the Member States." It's interesting that last month, the Chief Product Officer at OpenAI, former Chief Research Officer of OpenAI and Chief Technology Officer of Meta were directly commissioned into the US Army as Lieutenant Colonels, which implies a growing relationship: https://thegrayzone.com/2025/06/18/palantir-execs-appointed-colonels/
Joe Marler
2025-07-12 15:59:15 +0000 UTCThink the new unlock that multimodal LLMs enable for this purpose is essentially that they are able to work with unstructured Data ... text, images, video and audio out of the box and in large quantities. Used to be the case that our traditional AI algos and approaches were perfectly capable to work with structure data (tables, numbers etc.), but increasingly harder to work with the other sources. Additionally that they can use them essentially together (= all tokens) whereas before you needed lots of data prep for specific data sources to convert them into comparable formats. And as you alluded to the capability now to infer intent "naturally" for now at human level and soon beyond in soon infiinite data is well the big impact. Thanks for highlighting the danger of self censorship as well as also the danger on the production side (propaganda) ... the generative side of AI. Both things together are gonna be the true challenge... infinite ability to gauge automatically sentiment/danger/opinions by any one mass of people together with the ability to generate messages/content to influence and steer basically at a 1 to 1 level.
Stefan
2025-07-12 15:52:16 +0000 UTCGreat video. The self censorship is already very real. As you pointed out, any non-US citizen who may want or has a US visa is subject to social media profile review. I literally am hesitant to even post this fact given I have loved ones applying for such visas.
Sean Devs
2025-07-12 14:18:20 +0000 UTCGreat video! Though the EU is not exactly a model citizen either, we're currently pushing for laws that would allow backports for law enforcement on all new computers and ones that would make it criminal to build end to end encryption that can't be intercepted by the provider if so requested by authorities.
zero
2025-07-12 13:44:47 +0000 UTCLLM's provide the targets, perhaps soon humanoid robots will provide the enforcement. The most entertaining outcome will be the most likely outcome, unfortunately.
John Merkowsky
2025-07-12 13:17:37 +0000 UTCOne's principles matter more, as their power increases. AI increasingly empowers us all, including the state. So now is as important time as any to focus on our principles. One's opinion on the morality of nuclear weapons doesn't matter for most people, but it does for Harry Truman.
Bob Rein
2025-07-12 12:49:15 +0000 UTC
