2025-09-04 14:30:25
Prompt Attacks Against LLM-Powered Assistants in Production Are Practical and Dangerous
#security #llm
https://arxiv.org/abs/2508.12175
Prompt Attacks Against LLM-Powered Assistants in Production Are Practical and Dangerous
#security #llm
https://arxiv.org/abs/2508.12175
While I do maintain that "it's coming from the LAN" is not a good #security boundary, there are services where it is practical (eg. media center volume control), but also fault prone (oups my phone just switched to LTE for power saving – a generally justified thing).
Before I start formalizing how "a device can retain permissions it gets from being local for a few days&quo…
Harley is keeping a sharp lookout for sharks while our granddaughter is digging on the beach.
#DogsOfMastadon #BeachLife #Security
Hey #dotnet folks and #security wonks, join our #livestream today to learn about FAPI 2.0 and how to enhance security at your organization with the latest specification.
Also, drop in and say h…
Excellent article by @…, given a timely illustration by AWS's problems this morning.
"... this year, multiple important companies not just in the UK but globally and ranging from manufacturers to retail stores have been knocked almost completely out of production by cyberattacks. ...
"... British businesses have to start doing more to secure their IT systems and to create plans for how to keep running if something happens – potentially with plans to run systems without networked computers or with pen-and-paper backups if required. ...
"In 2019 we knew that a future pandemic was inevitable at some point but the lessons from previous pandemics and pandemic wargame exercises had not and still have not been fully implemented. ...
"Since then, the foreseeable disasters like pandemics, climate change or malicious hackers have been joined by another one – a hostile government that actively controls our tech sector."
#security #cybersecurity #AWS
20 ans après avoir été traumatisé par le projet Palladium de Microsoft, je m'intéresse enfin au TPM2
#Security
I disabled my browser password extension for now #Security
Passwords are a scourge foisted on elderly tech-users. :(
I just got back from a house call with a fellow who can’t remember his passwords for his email. None of his backups work. His little book of passwords, useless.
There is no support to call for help.
His only option is to create a new email and leave the other behind.
Over and over I have seen this trouble. Not always to this extent, but always the same frustration.
Passwords are elder abuse.
#techsupport #life #security
Are you worried your #dotnet #security could be more secure? Join us for a #livestream on August 21st, 2025, to discuss FAPI 2.0, its relation to
Want to know how to write and distribute #SecurityAdvisories that can be parsed and processed automatically?
Freshly announced are this years workshops for the Common Security Advisory Framework (#CSAF). They will be held in Nuremberg, Germany, November 10th to 12th.
See
Well done for making us all safe online:
#onlinesafety #security
#Schlagzeilen, die ich nicht lesen möchte:
#Security #Hacker #dataleak
Good post from @…:
"Trying to address the real issues going on in tech can’t rely on shaming average users for not conforming to an imagined version of reality that doesn’t exist and for not “just” doing things that aren’t really viable in light of everything else they’ve got going on."
#systems #tech #privacy #security
#UK Police Investigating #Handicapped Entrance #Security Scam At #Wembley After Hundreds Reportedly Gained Entry On Same
Here are some key takeaways from implementing #PyPI attestations in #Gentoo:
• With OpenPGP, you need to validate the authenticity of a key. With attestations, you need to validate the authenticity of the identity (i.e. know the right GitHub repository). No problem really solved here.
• They verify that the artifact was created by the Continuous Deployment workflow of a given repository. A compromised workflow can produce valid attestations.
• They don't provide sufficient protection against PyPI being compromised. You can't e.g. detect whether new releases weren't hidden.
On the plus side, TOFU is easier here: we don't have to maintain hundreds of key packages, just short URLs on top of ebuilds.
Security-wise, I think PEP 740 itself summarizes it well in the "rationale and motivation" section. To paraphrase, maintainers wanted to create some signatures, and downstreams wanted to verify some signatures, so we gave them some signatures.
#security #Python
I've drafted support for verification of #PyPI provenance for #Gentoo.
You know, the new fancy thing that protects against supply chain attacks on PyPI, and verifies that you're using genuine #GitHub artifacts. Because, you know, GitHub repositories and deployment pipelines are an unlikely attack vector. And you definitely don't need to worry about #Microsoft owning the keys, the repositories and the pipelines at all.
#security #Python #SigStore
Please do your part! This page makes it easy, just fill in the details, and send the email. I did it, so should you! This same shit is happening all over again. Wtf?
#ChatControl #EU #europe #FightBack #privacy #FightChatControl #security
Well, I am complaining about #AI slop introducing some random bugs in a minor userspace project, and in the meantime I learn that #Linux #kernel LTS developers are using AI to backport patches, and creating new vulnerabilities in the process.
Note: the whole thread is quite toxic, so I'd take it with a grain of salt, but still looks like the situation is quite serious.
"You too can crash today's 6.12.43 LTS kernel thanks to a stable maintainer's AI slop."
And apparently this isn't the first time either:
"When AI decided to select a random CPU mitigation patch for backport last month that turned a mitigation into a no-op, nothing was done, it sat unfixed with a report for a month (instead of just immediately reverting it), and they rejected a CVE request for it."
#security #LLM #NVIDIA #Gentoo
Do you use #Orbot?
#Android #GrapheneOS #Privacy #Security #Tor #TorBrowser #Google #Apple #iOS