George Santos given seven-year prison term for fraudulent congressional run
(How long until Trump pardons him?)
https://www.theguardian.com/us-news/2025/apr/25/george-santos-republican-sentenced
"what is it you want in return, Carnifex?"
"It must be some kind of weird reality bleed-through effect"
used to be https://twitter.com/jonpsp
I drink beer: https://untappd.com/user/jpennycook
Sometimes I buy music: https://bandcamp.com/jpennycook
I even see bands live: https://www.setlist.fm/user/jpennycook https://www.last.fm/user/jpennycook/events
I even edit maps from time to time, but I got bored with it: https://www.openstreetmap.org/user/jpennycook
last.fm (music I've listened to) | https://www.last.fm/user/jpennycook |
MusicBrainz | https://musicbrainz.org/user/jpennycook/edits/open |
Rambling Readers (books I've read) | https://ramblingreaders.org/user/jonpsp |
https://www.linkedin.com/in/jpennycook/ |
Tabletop scenario for you:
Employee gets into a dispute with employer, leaves, had sensitive role. Employer revokes access, devices etc. Employee had logged in via BYOD to email, IM etc.
Due to Recall, employee walks away with 6 months of screenshots of everything she's ever worked on in a text indexed form - every email, chat, document, Teams call with video snapshots, transcripts of verbal calls etc - even if they set M365 to not store documents locally.
What does the employer do now?
George Santos given seven-year prison term for fraudulent congressional run
(How long until Trump pardons him?)
https://www.theguardian.com/us-news/2025/apr/25/george-santos-republican-sentenced
The request, spelled out in an email obtained by ProPublica, comes amid concerns that DOGE has overstepped its bounds in seeking highly restricted private information about taxpayers, public employees or federal agencies.
Researchers at HiddenLayer have developed a prompt injection that successfully bypasses instruction hierarchy and safety guardrails across ChatGPT, Gemini, Copilot, Claude, Llama, DeepSeek. Qwen, and Mixtral. It bypasses
model alignment and produces outputs that violate AI safety policies. The outputs concern Chemical, Biological, Radiological, and Nuclear; mass violence; self-harm; and system prompt leakage.
I'm still on the fence about whether these sorts of prompt injections are newsworthy. I mean, how valuable the information they yield is already easily found on the Internet. Anybody want to weigh in?
https://hiddenlayer.com/innovation-hub/novel-universal-bypass-for-all-major-llms/