New Chinese AI models (Kimi K2) trained on the cheap (for few millions).

According to one scenario of Reserve Bank of Dallas, AI superintelligence leads to GDP=0 (all humans are dead). www.dallasfed.org/research/eco...

Fond memories of 2016-2019 GDPR+ePrivacy marathon, technical+policy. Here we go again? Would you like me to do similar deep dives and engagements (see eprivacy/gdpr tags on my blog ;-) )?

The leaked #GDPR modification is indeed quite strong. Total allowance for training AI on personal data, without consent or awareness of the person.

My comment in @Telegraph on AI propaganda. When it happens through AI/LLM models it may well be an info-laundry. Instead of reading a dubious site, you see it in a 'respected' AI output. Today, millions of people trust chatbots more than sources. www.telegraph.co.uk/business/202...

Malware is using AI to dynamically generate malicious scripts, obfuscate their own code to evade detection, and leverage AI models to create malicious functions on demand, rather than hard-coding them into the malware

Science has a problem. The integrity of research is increasingly threatened by a deluge of fabricated publications. Estimates suggest that in some fields, as many as 5–15% of all papers may be fake. Artificial intelligence has further accelerated this practice.

Switzerland will let licensed non-bank firms issue Swiss-franc stablecoins that are fully backed by segregated franc reserves. Anti-money-laundering rules allow blocking, freezing, or reversing transfers and limit use to verified users.

My comment in @Telegraph on AI propaganda. When it happens through AI/LLM models it may well be an info-laundry. Instead of reading a dubious site, you see it in a 'respected' AI output. Today, millions of people trust chatbots more than sources. www.telegraph.co.uk/business/202...

The US conducted a test of the Minuteman III intercontinental ballistic missile, a system designed to carry nuclear warheads. The missile flew 6,760 km to the Pacific. The test confirmed the equipment remains operational. www.vandenberg.spaceforce.mil/News/Article...

Sorry to disappoint you.

This month, European Commission is to propose rules relaxing privacy/data protection. On the table: letting easy AI training on user data without consent, relax tracking consent. (Though reducing consent fatigue is a good goal)

Do people have welfare responsibilities toward AI? Anthropic now argue that models might have morally relevant preferences or experiences, “interests.” Such systems lack free will and consciousness, why would such considerations be relevant?https://www.anthropic.com/research/deprecation-commitments

The European Commission plans to modify the #GDPR to make it much easier to train AI on private personal data. Without users' consent.

Arxiv, the leading repository for scientific preprints has stopped accepting certain computer science papers (review/survey) due to the flood of AI-generated JUNK. Only works that have already undergone peer review in a journal or conference are acceptable blog.arxiv.org/2025/10/31/a...

I’m giving a talk at King’s College London @kingscollegelondon.bsky.social @warstudieskcl.bsky.social on current developments in information/cognitive warfare, and the future. Fully automated influence operations are no longer hypothetical. How to defend?

AI-enabled browsers can be used to exfiltrate data and execute actions on user behalf. Expect session abuse and silent leaks, eventually? Also, nice mixing of text with commands. Here we go again. techletters.substack.com/p/techletter...

How to defend industrial control systems (i.e. in energy sector) with the assumption that a vendor may become compromised, or malicious, at some point? How to vet for patches? Are there any standard reverse engineering/bindiffing patterns or approaches? Is it being done by someone?

So, do you know what the problem is? 🤣

My newsletter #TechLetters ☕️ AI browsers can be hacked with words. CIA cyber-hit Venezuela, and what. HackingTeam returns. Conti operator in court. RAND calls for US-Chuna cyber détente. techletters.substack.com/p/techletter...

I’m giving a talk at King’s College London @kingscollegelondon.bsky.social @warstudieskcl.bsky.social on current developments in information/cognitive warfare, and the future. Fully automated influence operations are no longer hypothetical. How to defend?

AI-enabled browsers can be used to exfiltrate data and execute actions on user behalf. Expect session abuse and silent leaks, eventually? Also, nice mixing of text with commands. Here we go again. techletters.substack.com/p/techletter...

Those who control the process of disinformation must have access to information. Those who manipulate information rule the world.

Ukrainian Oleksii Lytvynenko, extradited from Ireland, appeared in U.S. court. He was to help deploy Conti ransomware: network intrusions, data encryption, ransom demands, threats to leak. www.justice.gov/opa/pr/ukrai...

My comment on @theverge.com. Today, AI-controlled web browsers introduce security and privacy risks. It's also kind of a step backwards, are we returning to text-based browsers (also walled gardens, competition protection, anybody)? ;) www.theverge.com/report/81008...

The past five years have shown that some “cyber norms” are inadequate, outdated and unsustainable, essentially out the window. It’s time to restart the process with a more realistic approach. What began as wishful thinking had clearly unraveled by 2019–2022. Sorry.

I'm exploring new engagements! If your team needs expertise in cybersecurity, risk assessment, tech policy, regulations (GDPR, etc.), tech standards, strategic insight, or Comms/PR, let's talk! Open to contract, or flexible roles. DM/email at [email protected].

CIA was making successful cyberoperations/attacks against Venezuelan government (disabling the computer network used by Venezuelan leader Nicolás Maduro’s intelligence service). The goal apparently was to "do something". edition.cnn.com/2025/10/29/p...

Looks like there's no AI bubble, right? Nvidia's entire Blackwell chip production for 2025 has been sold out. J. Huang announced that Nvidia has received over $500 billion in orders for delivery through 2026.

My comment for De Standaard. Poisoning LLMs with tainted information or propaganda to produce tainted replies is possible and not so difficult for a resourced threat actor. www.standaard.be/buitenland/r...

Chinese Communist Party has called for nationwide mobilization to achieve breakthroughs in semiconductors and other critical technologies. The stake is technological self-reliance.

Remember HackingTeam? They're back as Memento Labs. Their tools were used vs media, universities, government, financial institutions in Russia. Phishing + Chrome 0-day exploit. Just clicking a link was enough for full infection. Quite a disclosure. securelist.com/forumtroll-a...

My comment for @wired.com. LLMs become go-to reference tools, so information adversaries & digital propaganda actors will target this info infrastructure. Sanctioned-source citations in chatbots are a EU/US risk. It’s also a matter of compliance. www.wired.com/story/chatbo...

Employees use AI to generate fake expense receipts to deceive their employers, creating ultra-realistic documents in seconds. The goal is to claim reimbursement for expenses never incurred, exploiting how easily chatbots can now forge convincing receipts. www.ft.com/content/0849...

RAND advises to limit USA-China hot competition & stabilise based on peaceful coexistence & mutual understanding. They advise to negotiate cybersecurity treaties. www.rand.org/content/dam/...

And here’s a startup offering influencing/propaganda services. Called it ;) www.404media.co/a16z-backed-...

Good points? What's your take on this? www.ft.com/content/519a...

The most ridiculous demonstration would, of course, be a webpage which (to simplify) just makes the AI agent running the web browser make it leak all user data. Good luck!

AI-powered web browsers introduce new classes of security and privacy vulnerabilities. New ways to steal private information, loose reputation, and money. Uses in offensive cyber and propaganda are also imaginable.

💥Poland arrested 8 suspected saboteurs on October 21, 2025. Among them: a 21-year-old Ukrainian and two accomplices caught in Romania. They worked in a Warsaw warehouse. Explosives were seized.

OpenAI is offering increased ChatGPT usage for 7 days if the user makes the web browser the default. This is aimed at Chrome, not Perplexity's Comet.

In traditional browsers, everything stayed local. Your activity was yours. Web browser used to be the user’s own gateway to the web, acting for the user, not as a data source about them. Are AI-powered web browsers the only kind whose usage patterns could actually be subpoenaed?

In July 2025, the U.S. National Nuclear Security Administration (NNSA) and its Kansas City National Security Campus, which produces most of the components for America’s nuclear weapons, fell victim to a cyberattack.

Yes, another global IT malfunction which impacted this and that, and indirectly even more. Let's get used to that. The internet is no longer that decentralised. It's a question of risks, and costs. Regulations allow taking such risks. Prose of life.

Hundreds of classified military documents concerning 8 RAF and Royal Navy bases reportedly stolen – Russian hackers hacked British contractor Dodd Group, which provided services to the Ministry of Defence.

Signal is also down due to Amazon AWS outage. So yeah, secure and private communication is also dependent on reliable infrastructure.

Huge outage of Amazon/AWS. Cloud services affected. And many, many dependent websites/systems or services are impacted. A difficult time for Amazon engineers!

My #TechLetters ☕️ JLR cyberattack halts production. South Korea breached for 3 years. NSA vs China’s time system. AI poisoning made easy. techletters.substack.com/p/techletter...

Chinese authorities allege a 27-month NSA operation against the China's National Time Service Center. Initial credential theft apparently via a smartphone-messaging exploit. www.cert.org.cn/publish/main...

The success of passive resistance rests largely on its ability to secure widespread compliance in society. A government cannot be robbed of the popular support upon which it depends if only a few individuals act. A boycott, for example, requires participation by great numbers.