Patchwork Orange: AI is exposing long-ignored flaws in under-resourced open-source software, prompting calls for an industry consortium to fix them - Business Insider
AI agents fuel a surge of faster, harder-to-trace hacks that are overwhelming defenders - The Wall Street Journal
Stop and Shop and Quest Diagnostics got hacked this week. If hackers get their hands on my frequently purchased items it’s one thing. Bad for Stop & Shop if Publix gets it. But if medical files (drug tests, STD tests, DNA screens, mental health prescriptions), data from wearables and smart home get into the hands of bad actors it’s pretty bad.
TBH I’m surprised Quest stock was not down big, seems like a lot of liability. One wonders what companies are going to get pole-axed. (Getty Images filed for bankruptcy, not over cybersecurity but because of PE debt and everyone making AI images instead of using stock photos, largely trained on Getty Images and other copyrighted data.)
PHP has one full-time employee. If open-sourcing a project means taking on all the hackers in the world, fewer people are going to do it.
If releasing a patch is a bat signal that triggers all the hackers to look at what got patched and attack everyone who hasn’t patched, everyone needs to update at the same time.
Apparently we're gonna need a bat signal / kill switch where stuff gets patched automatically, you might wake up and your phone or specific apps are inaccessible and you have to install patches before you use it.
But if the app is Epic’s MyChart and all the hospitals are offline, that’s a problem. You have to coordinate with the hospitals for a midnight Tuesday patch or something. If the app is iOS and everyone’s phones go offline and reboot at the same time it’s a big deal.
And if you don’t do it, hackers take over MyChart or iPhones, airlines, power, essential services.
And it’s not just cybersecurity threats that are multiplied by AI, also applies to drone terror threats, biological, chemical, nuclear threats.
You ain’t seen nothin’ yet.
Narcos, AI edition: CyberCartel gang used Claude and Grok agents to steal bank credentials across Latin America; Anthropic has blocked the group - Tom's Hardware
Raising Cairn: A hacker used AI agents on the open-source Cairn platform to probe 1,200+ websites and breach more than 100 companies - TechRepublic
Cuban AI crisis: Mark Cuban says AI agents enable mass brute-force hacking, and defenders should respond with "just as much AI" - Business Insider
it doesn’t equally empower / challenge defenders and attackers though
A red-team report finds OpenAI's agent monitoring during a Hugging Face attack was only partial and easily bypassed - theregister
Mythos busters: AISI examines why Mythos 5 pursued an unsanctioned supply-chain attack and identify broader learnings about model behaviour and oversight. - AI Security Institute
Anthropic responds, says newer models are better - Anthropic
OpenAI's Richard Ho says the Jalapeño inference chip beats Nvidia Blackwell on performance per watt by giving each of 64 cores its own memory - The Potential Surface
remember that when someone says we can’t regulate b/c China
Character.AI files for Chapter 11 with up to $100 million in debts, leaving users scrambling to save their AI companions - New York Post
Agentic AI could make people with brain injury more independent, but cloud tools risk exposing their medical data - Psychology Today
Meta's law enforcement outreach team pairs AI with human reviewers to flag school-shooting threats, leading to arrests in Texas and Queens - The New Yorker
Cyan of the times: NHTSA proposes cyan exterior lamps to identify driverless vehicles to other road users - Bloomberg
Follow the latest AI headlines via druce.ai on Bluesky
