iiinova — where your digital self keeps evolving

Lukasz Olejnik @lukaszolejnik.bsky.social · 29/09/2026

I'll be speaking at the Brussels Privacy Symposium on 13 October. “To Share or Not to Share? Platform Regulation Meets Privacy”. Looking forward to discussing the tensions between competition law, platform regulation, and privacy. What's more important? Can we find a balance? #BPS2026

083

Reposted by Lukasz Olejnik

Gadi Evron @gadievron.bsky.social · 27/09/2026

An updated “Felony Bench” meme considering latest news (found on X by @lukaszolejnik.bsky.social)

063

Lukasz Olejnik @lukaszolejnik.bsky.social · 24/09/2026

OpenAI agents hacked Australian health service government systems. In June a model researching public medicine spending hit repeated access blocks on the Medicare statistics portal. www.pm.gov.au/media/press-...

1188

Lukasz Olejnik @lukaszolejnik.bsky.social · 23/09/2026

UK will create a National Centre for Information Defence to detect, attribute and disrupt foreign state information operations, attacks, and interference. Early days. The Centre is just announced so it has no shape, director, budget, headcount, launch date or legal basis yet.

1111

Lukasz Olejnik @lukaszolejnik.bsky.social · 21/09/2026

My #newsletter #TechLetters ☕️. OpenAI security was a text string. Claude hacked OpenAI. Russia’s election system leaked 111M voters. AWS lost a region to war. AI hallucination nearly triggered military action? techletters.substack.com/p/techletter...

techletters.substack.com

TechLetters ☕️ OpenAI security was a text string. Claude hacked OpenAI. Russia’s election system leaked 111M voters. AWS lost a region to war. AI hallucination nearly triggered military action?

Security

052

Lukasz Olejnik @lukaszolejnik.bsky.social · 18/09/2026

UK government is now advising civilians to prepare for major disruption, including the possibility of war. What do you think?

251

Lukasz Olejnik @lukaszolejnik.bsky.social · 13/09/2026

Proportionality weighs a restriction against the harm it prevents. If you seriously accept the "AI KILLS ALL HUMANS / WORLD DOOM" risk model, the harm is infinite. Absolutely no restriction is ever disproportionate.

1101

Lukasz Olejnik @lukaszolejnik.bsky.social · 13/09/2026

1) Open-weight models are improving with respect to closed frontier models. 2) AI Safety arguments as framed, may help an intervention to keep control.

192

Lukasz Olejnik @lukaszolejnik.bsky.social · 12/09/2026

My rough estimate is that around ~9 autonomous AI hacking incidents remain undisclosed/unknown. At least based on the scarce data points we have. techletters.substack.com/p/9-autonomo...

074

Lukasz Olejnik @lukaszolejnik.bsky.social · 12/09/2026

OpenAI's own AI agents again broke into someone else's servers. The victim was RubyGems, the registry millions of programmers download Ruby code from. www.rubyhack.ai

1146

Lukasz Olejnik @lukaszolejnik.bsky.social · 11/09/2026

Houthis used Claude to code ballistic missile guidance. The test flopped but their offensive didn't - they've seized the Red Sea gateway Saudi oil uses to dodge Hormuz, are blocking ships and may have hit the oil pipeline. Oil is near $110. Does this sound crazy? (true story)

174

Lukasz Olejnik @lukaszolejnik.bsky.social · 11/09/2026

Someone in Yemen used Claude AI to write the software for missiles. Three kinds, in fact. A ballistic missile with a range of 2,000 km, a family with a hypersonic glide variant, and a guided rocket. www-cdn.anthropic.com/e50be2e51e76...

01111

Lukasz Olejnik @lukaszolejnik.bsky.social · 10/09/2026

Some keep asking for the most likely scenario in which AI kills all humans and destroys civilisation. It’s important to understand that IT IS NOT POSSIBLE to identify it in advance. If we knew the route, we could close and solve it today.

3152

Lukasz Olejnik @lukaszolejnik.bsky.social · 10/09/2026

Here's how we'll protect humanity from extinction-level AI. We're safe between 2026 and until July, 2030. huggingface.co/security.txt

093

Lukasz Olejnik @lukaszolejnik.bsky.social · 10/09/2026

I'm open to consulting, full-time or part-time work across research, engineering, security, AI, regulation and policy. AI-first work very much in scope. This can include commissioned research, security/privacy review, AI & agent-system, technical analysis, policy/regulatory work me@lukaszolejnik.com

024

Lukasz Olejnik @lukaszolejnik.bsky.social · 10/09/2026

The new iPhone Fold is pretty cool. Now everyone on the subway or tram can see what impressive and awesome books you’re reading. (Though I suggest ordering a hardcover)

1152

Lukasz Olejnik @lukaszolejnik.bsky.social · 09/09/2026

The fact that ~anybody with a strong AI model and tokens can reprouce/redo/steal ~any idea from descriptions is a big argument against transparency and publishing supplemental data linked to research papers. This may effectively mean revealing internal secrets. Where this leads?

140

Reposted by Lukasz Olejnik

Raphael Satter @raphae.li · 05/09/2026

So more than 24 hours after we scooped the news that OpenAI’s agents had hijacked a small German wiki site and commandeered it as a comms platform, OpenAI acknowledges what it calls a “wiki incident.” www.reuters.com/business/med...

reuters.com

OpenAI acknowledges 'wiki incident' and need for more transparency around unintended AI behavior

OpenAI said incidents including a swarm of its agents hijacking a German site and using it as a springboard for cheating and other rogue behavior require more transparency.

56037

Lukasz Olejnik @lukaszolejnik.bsky.social · 04/09/2026

My comment for Reuters about the OpenAI-agent incident where agents apparently coordinated emergently during a routine web search and analysis task, probing a website for vulnerabilities and exhibiting exploitation attempts, persistence, and coordination. www.reuters.com/world/europe...

reuters.com

EXCLUSIVE: OpenAI agents hijacked German website in previously undisclosed AI breakout this spring

A swarm of rogue OpenAI agents hijacked a German website this spring and transformed it into a bulletin board for other AI agents, according to new research published Friday and two people familiar w...

2159

Lukasz Olejnik @lukaszolejnik.bsky.social · 03/09/2026

"When an experimental OpenAI model recently escaped its sandbox and broke into another company’s servers, OpenAI said it took its responsibility to prepare for such incidents “seriously” but described the incident as a technology/industry problem - not an OpenAI problem" www.ft.com/content/115c...

165

Lukasz Olejnik @lukaszolejnik.bsky.social · 26/08/2026

I broke lattice e-voting with GPT 5.6 Sol ;) arxiv.org/pdf/2608.21986

001

Lukasz Olejnik @lukaszolejnik.bsky.social · 26/08/2026

An off-the-shelf VM is not enough to contain a modern, cyber-capable AI agent. AI agent escaped a VM three times. First via known bugs, then via missed patches, then it found 0-days and chained them. blog.trailofbits.com/2026/08/26/v...

1198

Lukasz Olejnik @lukaszolejnik.bsky.social · 26/08/2026

Enter AI GRINDING to break cryptography! What happens when cryptanalytic ideas become cheap? AI agents allow efficient and mass-scale verification and breakage of cryptographic schemes and protocols. arxiv.org/abs/2608.21986

044

Lukasz Olejnik @lukaszolejnik.bsky.social · 25/08/2026

We broke a CRT-RLWE fully homomorphic encryption scheme and identified a key & plaintext recovery attack against this recently published FH encryption. The basic problem is surprisingly simple. Noise that is supposed to hide secret information disappears. eprint.iacr.org/2026/1734

150

Lukasz Olejnik @lukaszolejnik.bsky.social · 24/08/2026

Netherland's and French data protection agencies fined Uber 824 990 000 euros for automated driver blocking. Regulators said Uber let software deactivate drivers without meaningful human review, cutting off income and violating #GDPR rules on automated decisions. www.cnil.fr/fr/decisions...

cnil.fr

Décisions automatisées : sanction de près de 825 millions d’euros à l’encontre d’UBER

UBER regroupe UBER B.V., société néerlandaise située à Amsterdam, et UBER TECHNOLOGIES INC., société étatsunienne, dont le siège social est à San Francisco. UBER édite notamment une plateforme mettant...

074

Lukasz Olejnik @lukaszolejnik.bsky.social · 24/08/2026

My #newsletter #TechLetters ☕️ AI accelerates attacks on industrial control systems. Pro-Russian hackers expose Spanish security personnel. Data of 68% Latvian and 50% Poland (medical records) population leaked. techletters.substack.com/p/techletter...

064

Lukasz Olejnik @lukaszolejnik.bsky.social · 23/08/2026

Cyberattack on a small UK gas-fired power plant (~15MW) reportedly shut it for 4 days. Negligible for a national grid measured in tens of GWs, but strategically significant. It kind of shows that lightly protected, remotely operated energy assets can be disrupted by hackers.

185

Lukasz Olejnik @lukaszolejnik.bsky.social · 23/08/2026

Two cryptography papers coming soon. A dozen or so schemes and constructions didn’t survive close analysis: encryption, signatures, electronic voting, hash functions, and a few other things. Great fun, and the approach turned out to be quite systematic :)

141

Lukasz Olejnik @lukaszolejnik.bsky.social · 22/08/2026

Prompt engineering may be the first profession to fall victim to AI.

3191

Lukasz Olejnik @lukaszolejnik.bsky.social · 21/08/2026

Remote code vulnerability found in Unitree robots and the scary part is that the exploit is wormable. One compromised robot could infect other vulnerable robots nearby. Could attackers then take control of entire fleets of humanoid robots? boschko.ca/unitree-go2-...

1168

Lukasz Olejnik @lukaszolejnik.bsky.social · 21/08/2026

My comment in @reuters.com on a “test” of AI that crossed a remarkable line. AI agent attempted a real software supply-chain attack, then used deception against a developer who caught it. This crossed the line from autonomous hacking to interactive deception. www.reuters.com/world/how-te...

02214

Lukasz Olejnik @lukaszolejnik.bsky.social · 20/08/2026

Cyber threat actors are using AI to accelerate attacks on industrial control systems. Malicious PLC writes might eventually alter process or safety logic, move equipment outside safe operating limits, or even cause danger to humans media.defense.gov/2026/Aug/18/...

077

Lukasz Olejnik @lukaszolejnik.bsky.social · 19/08/2026

The data included names, photographs, phone numbers, email addresses, home addresses, contacts, and, in some cases, details about their units or roles. It is unclear whether the group hacked official systems or compiled the data from previously leaked, compromised, or publicly available sources.

067

Lukasz Olejnik @lukaszolejnik.bsky.social · 19/08/2026

Spain investigating a major data exposure linked to the pro-Russian hacker group NoName057, which released a 499-page document containing sensitive personal information on around 1,000 police officers, Civil Guard members, military personnel, security-related individuals. elpais.com/espana/2026-...

elpais.com

La Policía investiga el acceso del grupo ‘hacker’ prorruso NoName057 a datos sensibles de un millar de agentes y militares

Esta red, entre cuyos cabecillas está un profesor español, ha lanzado en los últimos años más de 500 oleadas de ciberataques

034

Lukasz Olejnik @lukaszolejnik.bsky.social · 19/08/2026

A 25-year-old fmr Goldman Sachs financial analyst in the U.S. entered detailed plans to murder his ex-girlfriend into ChatGPT, including places she frequented and how he intended to kill her. OpenAI flagged the conversations and reported them to the FBI. He was sentenced to eight years of probation.

2137

Lukasz Olejnik @lukaszolejnik.bsky.social · 18/08/2026

Just because a scheme or protocol is advertised as post-quantum doesn’t make it post-quantum, or even non-quantum secure. “Post-quantum” is a nice selling point, though. Very en vogue in 2026. The „we used pen and paper to attack…” is less so. 😉

051

Lukasz Olejnik @lukaszolejnik.bsky.social · 17/08/2026

Submitted a paper to Journal of Cryptology ;-)

060

Lukasz Olejnik @lukaszolejnik.bsky.social · 17/08/2026

How about disagreeing with this @financialtimes.com take? Adopting open-weight models does not equate adopting standards and governance. The models can be modified anyway so whatever “standards” / “governance” are baked in may be bent, cut out, replaced. That’s not how technological influence works

2102

Lukasz Olejnik @lukaszolejnik.bsky.social · 17/08/2026

My #newsletter #TechLetters ☕️ McEliece post-quantum crypto gets a A PUNCH IN THE FACE? US creates cyber-privateers. Cyber norms face a stress test. French tax office breach exposes 678,000 people. techletters.substack.com/p/techletter...

techletters.substack.com

TechLetters ☕️ McEliece post-quantum crypto gets a A PUNCH IN THE FACE? US creates cyber-privateers. Cyber norms face a stress test. French tax office breach exposes 678,000 people.

Security

010

Lukasz Olejnik @lukaszolejnik.bsky.social · 14/08/2026

French Constitutional Court struck down the social media ban for under-15s, ruling that it would effectively require age verification for every user, including adults.

1176

Lukasz Olejnik @lukaszolejnik.bsky.social · 14/08/2026

This is what Heritage Foundation also advised the US to do. Engage cyber privateers.

3104

Lukasz Olejnik @lukaszolejnik.bsky.social · 14/08/2026

New powerdul Chinese open source model GLM-5.3 refined to have powerful cybersecurity/cyberattack capability. They claim that the oldest found cybersecurity bug is 45 years old, from 1981. z.ai/blog/glm-5.3

z.ai

175

Lukasz Olejnik @lukaszolejnik.bsky.social · 13/08/2026

My comments in @financialtimes.com about USA authorizing private firms to conduct offensive cyber operations/attacks, including with physical destruction effects (but not lethal). Licence to hack and destroy. Risk of interstate conflict goes up. www.ft.com/content/a468...

093

Lukasz Olejnik @lukaszolejnik.bsky.social · 13/08/2026

USA is building a state-controlled private cyber force. It lets vetted US companies conduct covert access, surveillance and cyberattacks against foreign criminal networks + physical destruction. For lethal effects, there's a stop.

055

Reposted by Lukasz Olejnik

Lukasz Olejnik @lukaszolejnik.bsky.social · 12/08/2026

Platforms, public institutions and security teams need ways to examine that continuing process and test their defences against it. The information operation campaign is now an experimental object. Full description: techletters.substack.com/p/we-ran-the... Blog blog.lukaszolejnik.com/how-we-built...

techletters.substack.com

We ran the first fully autonomous end-to-end information operation (TechLetters Insights)

We conducted the first fully autonomous information operation under controlled conditions.

023

Reposted by Lukasz Olejnik

Lukasz Olejnik @lukaszolejnik.bsky.social · 12/08/2026

In the AI age, the unit of analysis of information operations needs to be the campaign, not individual or even bunch of posts. An autonomous operation can persist across accounts, monitor its environment and alter its behaviour over time.

techletters.substack.com

We ran the first fully autonomous end-to-end information operation (TechLetters Insights)

We conducted the first fully autonomous information operation under controlled conditions.

132

Reposted by Lukasz Olejnik

Lukasz Olejnik @lukaszolejnik.bsky.social · 12/08/2026

Releasing a multi-month work today! We conducted the first fully autonomous information operation (under controlled conditions). It ran the campaigns end-to-end. It's can work in full autonomous mode. arxiv.org/abs/2608.10920

1188