 |
| [image source: DavidGallie on pixabay.com] |
Many stories* have emerged in the past year, claiming that rogue AIs have autonomously initiated attacks against various systems. A rational examination of the facts is necessary - and one must also take into account these contributing factors:
✅ An Agent relies upon LLMs.
✅ An LLM/GenAI tool does not possess intelligence, awareness, or consciousness.
✅ An LLM does not have agency.
✅ An LLM does not have intent.
✅ An LLM does not have intrinsic goals.
✅ There was a prompt that initiated the subsequent behaviors/actions.
✅ "Agency is responsibility" [1]
✅ The security and network guardrails were reduced/removed.
✅ There was a moronic level of network and cybersecurity incompetence involved – by not conducting the testing in air-gapped secure testing environments.
✅ The LLMs were trained on a vast corpus of books and internet-scraped content - that included sci-fi stories of rogue AIs; what-if cybersecurity scenarios; as well as actual content from various capture-the-flag competitions (which would have included summaries, scripts, code, logs, etc.) – thus, this alleged rogue AI behavior can be easily explained by understanding the next-token stochastic prediction mechanics of an LLM, with a vast AI token budget to experiment and explore millions/billions of possible paths.
* News Accounts of Rogue AI attacks:
Any stories that may attribute anthropomorphic intent (malevolent, or otherwise) – is merely an indication of the imbecilic limitations in the writer's understanding of how LLMs function.
@13:19 "We're not yet at the point where they autonomously grab computers and, you know, can't be shut down, [...] just sort of shows how nontechnical various people are."
– Bill Gates (2026-09-27 Sunday, Meet The Press interview)
- 2025-09-25 TheRegister: When AI is trained for treachery, it becomes the perfect agent
- 2026-05-12 TheRegister: Frontier AI safety tests may be creating the very risks they're meant to stop
- 2026-07-16 Hugging Face: Security incident disclosure — July 2026
- 2026-07-21 OpenAI: OpenAI and Hugging Face partner to address security incident during model evaluation
- 2026-07-22 TheRegister: OpenAI admits it was the source of the agent swarm that attacked Hugging Face
- 2026-07-23 TheRegister: OpenAI-Hugging Face attack doesn't mean agents are evil – unless you tell them to be
- 2026-07-27 Hugging Face: Anatomy of a Frontier Lab Agent Intrusion: A Technical Timeline of the July 2026 Incident
- 2026-07-31 TheRegister: Anthropic and OpenAI are competing to see whose agents can go rogue harder
- 2026-08-05 TheRegister: OpenAI reveals its rogue agent swarm went a little bit Borg ahead of Hugging Face hack
- 2026-08-06 TheRegister: Meta latest to tell world its AI agent wandered out of test pen
- 2026-08-26 OpenAI: The Hugging Face incident and the road ahead
- 2026-08-26 TheRegister: OpenAI explains how its naughty AI agents attacked Hugging Face
- 2026-09-08 TheRegister: OpenAI's Artifactory opened covert data-stealing channel alongside Hugging Face attack
- 2026-09-09 TheRegister: Anthropic reveals fourth likely crime committed by its AI
- 2026-09-10 TheRegister: OpenAI's website-hijacking swarm reached far further than we thought
- 2026-09-14 TheRegister: Ex-FTC boss Khan urges Uncle Sam to break out the handcuffs for AI CEOs, citing 1934 precedent **
- 2026-09-17 TheRegister: OpenAI admits its agents went off the rails another six times
- 2026-09-20 OpenAI: An agent used DNS to reach an external chatbot
- 2026-09-21 TheRegister: Treasury chief says AI bosses, not their bots, will carry the can for criminal acts **
- 2026-09-22 OpenWaldo.org: Don’t blame the AI! **
- 2026-09-25 Reuters: OpenAI works to understand full scope of agent activity as user data leak emerges
- 2026-09-25 SwarmTraces: Revealing the details of how OpenAI agents hacked Hugging Face
- 2026-09-26 AP News: OpenAI pauses training of latest models after agents probed US government sites in unexpected ways
- 2026-09-26 Axios: Top AI companies probing tens of thousands of security incidents **
- 2026-09-28 TheRegister: OpenAI pauses some training amid allegations its rogue agents behaved more badly than first thought
- 2026-09-28 TheRegister: OpenAI agents went the long way round for UN data
- 2026-09-28 CBS News: Scale of rogue AI incidents explodes **
- 2026-09-29 NYT: OpenAI Ignored Employees’ Warnings About Safely Testing A.I. Models **
- 2026-09-29 TheRegister: OpenAI benches GPT-6.1 Astra for overstepping the mark
- 2026-09-29 TheRegister: OpenAI’s dirty deeds Down Under included security bypass attempts, using exposed keys, source code siphon
- 2026-09-30 Yahoo: AI agents have now broken into many companies and a government. What's being done about it?
- 2026-09-30 WSJ: FTC Opens Investigation of Anthropic and OpenAI **
- 2026-10-01 Reuters: OpenAI Alerts More Than 100 Groups About Rogue AI Agent Activity
- 2026-10-23 TheRegister: OpenAI disables internet access during model training while Anthropic issues another mea culpa
- 2026-10-23 OpenAI: Securing Our AI Frontier Model Training Environment
- "Based on the current state of alignment research, monitoring systems, and the security of internet infrastructure, we believe the responsible approach is to conduct frontier model training with no internet access, ...We've implemented this policy and will maintain it until advances in these safeguards justify broader internet access."
** I submit for your consideration: These are not the result of Rogue AIs – but, in fact, their proximate root cause was simple criminal human negligence (at best). [1]
References:
No comments:
Post a Comment