Notices by Taggart :ifin: (mttaggart@infosec.exchange)

Pretty wild to see hacking content creators I once admired get completely AI-brained. For years you've made a name for yourself telling people the path to success was learning certain skills. And now, suddenly, the message is just "let the models do it?"

I know what the models can and can't do. But I also know that they require domain expertise to keep them on track, and there's only one way to develop it. Hint: it isn't writing prompts.

@winterknight1337 @SecureOwl

I'm coming to the conclusion that community-owned and operated small clouds (co-ops) with easy onramps for self-hosting open source services like mail, storage, and VPN are the only way forward. Every corpo service is eventually going to make you ashamed to use it.

Over here writing code line by line, comment by comment. You know, like someone who "hates technology."

Alphabet's first equity raise since initial offering should probably be bigger news.

https://www.semafor.com/article/06/04/2026/hyperscalers-are-strapped-for-cash

Fundamentally. Unsecurable.

You can't rid yourself of this vulnerability in LLMs.

https://www.safebreach.com/blog/gemini-voice-assistant-prompt-injection-exploit/

It really blows my mind that nobody is taking seriously the risk of processes built assuming the indefinite availability of AI compute—at current costs.

The financials don't bear that faith out. Even if the "industry" doesn't collapse outright, the costs will skyrocket, and then what will you do?

Hot diggity dog.

The briefing concludes that standalone generative AI systems, based on unlawful web scraping, depend on mass invasions of privacy by design, and are fundamentally incompatible with [International Human Rights Law]. As such, Amnesty International is calling for a prohibition of such systems, including where such systems are identified as exacerbating existing inequalities or creating new forms of discrimination.

https://www.amnesty.org/en/documents/pol40/0996/2026/en/

@kims Is there anything more anti-LLM than typing source code from the pages of a magazine?

RE: https://infosec.exchange/@ifin/116622369159743355

Gonna go back to sending software by mail until we can figure this out.

Lmao okay. Well the Claude/Claude Code "sandbox" does not prevent any DNS resolution either, soooo

https://www.theregister.com/security/2026/05/20/even-claude-agrees-hole-in-its-sandbox-was-real-and-dangerous/5243662

RE: https://infosec.exchange/@ifin/116605052950779161

I'm really hoping this, and the last few weeks, are a tipping point toward taking supply chain security seriously. Initial access appears to be a compromised VS Code extension.

Bluesky once again appears to be struggling

This entire report from the Ontario government on genAI systems is worth a read, but the review of healthcare scribe accuracy is pretty devastating, imo. This has to work for the tech to be worth anything. If the notes in the chart are wrong, the whole thing falls apart.

https://www.auditor.on.ca/en/content/specialreports/specialreports/en26/2026_AI_EN.pdf

If you think folks are going to let things like this get built in their communities and operate with impunity, you have another thing coming. I'm not saying they should be sabotaged; I'm saying they positively will be.

https://www.yahoo.com/news/articles/ai-data-center-bigger-2-124356503.html

Truly phenomenal work from @flyingpenguin here. A breakdown of the timeline of the Ubuntu attack, the business behind the booter service, and Cloudflare's problematic role.

https://www.flyingpenguin.com/can-someone-please-explain-whether-cloudflare-blackmailed-canonical/

So the Claude extension allows any other extension to inject JavaScript into claude.ai and run it? Cool cool cool.

Yeah, don't let this one in.

https://layerxsecurity.com/blog/a-flaw-in-claudes-browser-extension-allows-any-extension-to-hijack-it/

For real though, more schools should know about Gibbon: https://gibbonedu.org/

Do you see Moodles getting owned? I don't think so!

@cwebber Willison's distinction (and previous iterations of it) strike me as dangerously naive. Based on my own experience with both the tool and, y'know, human nature, nothing but the most draconian guardrails will prevent people from taking the easy way with these tools.

We were never going to do this safely.