Imagine testifying in court that you relied on a forensic tool known to make stuff up.

[mttaggart@infosec.exchange]

Imagine testifying in court that you relied on a forensic tool known to make stuff up.

Defense attorneys are gonna love this one.

https://www.autopsy.com/autopsy-4-23-0-release-claude-ai-assistant-mcp-cyber-triage-integration/

[mttaggart@infosec.exchange]

Uhhh also this tool is designed for proprietary, corpo, cloud-hosted models. Dunno about y'all but my forensic evidence can't leave the lab in any form without chain-of-custody procedures. What are we doing here?

[avuko@infosec.exchange]

@mttaggart

Can these children please move to a playground so the adults can work? Thanks!

[707kat@mastodon.art]

@mttaggart yup. And in EU public caseworkers gladly sidestep GDPR to use the now designated ‘meant for entertainment only’ copilot slopware to process private citizens most private and intimate data and details with the blessing and encouragement from their bosses.

#twitsforbrAIns

[luxliquida@critter.cafe]

@mttaggart Perhaps the worst part is that law enforcement is unlikely to care enough about its inaccuracy to avoid using it in the first place... It hasn't stopped them from using dodgy tools in the past (ShotSpotter comes to mind)

[swappage@infosec.exchange]

@mttaggart It looks to me that they don't care about LEA forensics analysis any more.

They are focusing more and more on enterprise databreach investigations and IR, so they probably don't care too much about where data goes.

still, I think having an AI jumping to conclusions is very very bad.

I wonder if using local models can potentially help in scraping huge evidence collections tho.

Having something that assists in heavy tasks like timeline vent correlation shouldn't be THAT BAD: human supervision is obviously a must.