Notices by Lukasz Olejnik (lukaszolejnik@mastodon.social)

OpenAI shipped a telemetry system that logs more than the actual work being done. Codex burning through SSDs at a rate of ~640 TB/year – one user hit 37 TB written in 21 days. On a consumer SSD
that’s full drive death in under a year. https://github.com/openai/codex/issues/28224

Cyberattacks are not attacks. Cyber weapons are not weapons. Cyberwar is not war.

Linux kernel security list is drowning in duplicate AI-flagged bugs; same issues, with same tools, but different names of human submitters. Maintainers have now formalized the obvious - AI-found bugs are public by definition. New docs define 5 failure modes for AI-assisted reports: too long, Markdown-heavy, threat model-ignorant, reproducer-free, patch-free. Non-compliant reports risk being ignored. Most AI-flagged issues aren’t even real vulnerabilities anyway. 500 submitters, 1 CVE, 0 patches?

My comment for WSJ. Trolling is now a standard tool of statecraft. Lego, Call of Duty, Grand Theft Auto or Nintendo work well as propaganda because they are universally recognised cultural cues with preloaded emotional associations. They are also perfect for fast-paced communication styles. Expect this new approach to diplomacy, signaling, international relations, and information operations to be used more frequently. https://www.wsj.com/world/lego-ai-propaganda-wars-iran-ab29cc6c

A flow-diagram for handling identifier/cookie consent. Entry checks, user choice, storage, withdrawal, multi-device sync, shared device protection. Warning: #GDPR / #ePrivacy Regulations may change this soon.

OpenAI's ChatGPT is getting support for displaying ads. Would you prefer it in a separate subwindow, or maybe embedded in answers? https://mashable.com/article/chatgpt-ads-leak

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

Cyberattack on Jaguar Land Rover results in a prolonged production shutdown. 25% of suppliers have already taken steps to pause production and temporarily lay off workers. The attack has now forced a complete halt until October 1st, costing the company over $1.36 billion in lost revenue and affecting 200,000+ jobs across the supply chain. This is one of the most devastating cybersecurity incidents in automotive history.

I’m thinking of making a presentation “Malicious use of AI, one year after” (for security and information operations), where I would show the current status quo, and what’s imminent (including based on my own tests/development). Any ideas where to submit it to?

AI vulnerability/bug founds and reports is a huge problem. Curl has banned the use of AI-generated submissions via HackerOne because none of it made any sense, and is a waste of resources and time. "We are effectively being DDoSed. If we could, we would charge them for this waste of our time" https://hackerone.com/reports/3125832

No cyber threat actor has been found to use any public AI/LLM for serious cyberattack/cyberoperation goals. Unless they're trolling, because it is already possible to such tools in completely undetectable ways. Cybersecurity and LLM providers have no visibility here.

@GossiTheDog So what was the question to the security operations? "Tell me if there's at least one Ukrainian IP address involved"?

@GossiTheDog But only a single country is mentioned, why?

Beware of narrative stretches used for PR purposes. Even a single packet sent from a Ukrainian IP in a DDoS attack on X platform doesn’t imply 'Ukraine behind it'. This is an oversimplification, perhaps to to divert attention. Attribution doesn’t work that way—an IP is inconclusive. And much of Ukraine is occupied, with traffic operated by Russian ISPs.

Malicious open source models are being uploaded to popular repository hugging face. This will be a completely new cybersecurity risk. Now it's merely code execution. But expect tainted/poisoned weights impacting outputs. Python reverse shell script enables remote command execution. On Linux, it spawns a `/bin/sh` shell, on Windows, it launches PowerShell and enables bidirectional communication. https://www.reversinglabs.com/blog/rl-identifies-malware-ml-model-hosted-on-hugging-face

Stealing passwords and PINs entered by Apple Vision Pro users. As you can see, you never know when a lecture on double integrals over surfaces might come in handy. https://arxiv.org/abs/2409.08122

My strategic privacy analysis. Is Google undoing a decade of progress on privacy? Their new policy allows invasive device fingerprinting for tracking user activity. Here’s my deep dive into what this means for privacy—and the future of AI. https://blog.lukaszolejnik.com/biggest-privacy-erosion-in-10-years-on-googles-policy-change-towards-fingerprinting/

Russian cyber threat actor Turla hacked 33 infrastructure nodes of Pakistani cyber threat actor to attack other targets, to deploy own cyber tools (malware) for cy-espionage purposes in the Middle East, like India. It delays attribution. What's the most vulnerable sensitive authorization in the world? "nation-state and cybercriminal endpoints and malware especially vulnerable to exploitation since they are unable to use modern security tools for monitoring access" https://blog.lumen.com/snowblind-the-invisible-hand-of-secret-blizzard/

Russian cyber threat actor is targeting Ukraine military conscript system, trying to infect conscripts systems (Windows, Android, macOS, iOS) with malware, and spreading narratives and content to undermine support for Ukraine's war mobilization. https://cloud.google.com/blog/topics/threat-intelligence/russian-espionage-influence-ukrainian-military-recruits-anti-mobilization-narratives

Russia issued a monetary fine on Google: 2 undecillion rubles ($2,500,000,000,000,000,000,000,000,000,000,000) after refusing to restore the accounts of pro-Kremlin and state-run media outlets. https://www.themoscowtimes.com/2024/10/29/russia-fines-google-25-decillion-over-youtube-bans-rbc-a86846