Gotta love Slack’s UI that keeps presenting this to me and no matter how many times I click the “Not now” it stays on the screen. Seriously, I don’t need a summary of “Sounds good” Hounding users to use AI features they don’t want or need is the new normal.

Join Nils Amiet and me at Black Hat USA this year for Hack to the Future. This isn't a prompt injection talk. This is about the hidden dangers of deploying these tools in your environments and the potentially devastating vulnerabilities that can result. https://www.blackhat.com/us-25/briefings/schedule/#hack-to-the-future-owning-ai-powered-tools-with-old-school-vulns-45871

@gsuberland Always :)

Trying to get your case dismissed by making a claim that your bot/product has First Amendment rights is an odd flex, and the way they had to word it to attempt to make legal sense blames the victim. Thankfully, the case wasn’t dismissed. https://scholarblogs.emory.edu/proflawrence/files/2025/05/Garcia-v.-Character-Technologies-Inc.-et-al-Entry-115.pdf

I'll be speaking at the AI Summit on August 5th. Looking forward to a great conversation.

Next time someone asks me if I know what hustling looks like, I'm gonna show them this. If you can play Denny's, you can play anywhere 😆

The next battleground is in sight, and things are going to move fast. Half-baked tech pitched as transformational will be quickly adopted and thrown in front of children without any validation, but the demos will be amazing!

Move fast and break things is all well and good until the thing being broken is your child. The administration signed an executive order mandating AI in K-12 education. The impact could be irreparable damage if not done right. I’ve written up some thoughts outlining the potential damage and making some recommendations. https://perilous.tech/2025/05/16/our-next-disaster-negative-impacts-of-mandating-ai-education-in-k-12/

Nonsense. The real AI risk isn't superintelligence, it's superincompetence. A sentient super AI deciding to wipe out humans because they are pesky is microscopic compared to some idiot hooking an LLM up to a safety-critical system because their boss tells them to use AI for everything.

It’s a dystopian ink kind of day.

It's surreal to see many of the things I've been warning about over the past couple of years, starting to come to fruition, where ever-present surveillance devices are pitched as your "friends" and university education is about nothing more than vibes.

The personal AI Safety attributes I outlined back in March of 2024 are even more important than ever. We need to ensure these systems are safe to use, ensuring Security, Privacy, Alignment, and Reliability.

https://perilous.tech/2024/03/06/introducing-spar-safety-attributes-for-personal-ai-assistants/

Dystopia is now 😎 https://futurism.com/ai-dead-brother-impact-statement

“Unexpected RCE” has to be one of the silliest names for a vulnerability I’ve heard in quite some time.

@msh Yes, he's saying the absurd part out loud and not listening to himself.

Everybody thinks AI will replace everyone else's jobs except their own. We continue to make the mistake of observing the nuance in our own jobs and assuming none in others.

AI-powered developer productivity tools are not trustworthy collaboration partners. They can leak your secrets, lie to you, make stuff up, and be convinced to take actions that aren’t in your best interest. Consider them adversarial and utilize additional security controls.

@lmorchard @JessTheUnstill This is true and extends to the ridiculous claims made about AI advancement. It's performance art for investors.

@RandomDamage @bradley You know it :)

I wrote some thoughts on adding sanity on the pace of AI advancements. The conversation shouldn’t be monopolized by people selling us stuff and influencers huffing vibes.
https://perilous.tech/2025/04/25/some-sanity-on-the-pace-of-ai-advancement/

@bradley The next phase will be “ludicrously focused” ads.