AI platforms and agents
EU KIDS Act asks platforms to prove child safety compliance.
The proposal introduces obligations that platforms must demonstrate, but proving when protections were active is left open.
Newsroom · Today in evidence
Today in evidence.
Where records were questioned, and what a witness would add. Today's briefing appears here once our editors have approved it.
AI platforms and agents
Australia's frontier AI rules ask for risk assessments and incident reports.
New obligations mirror EU AI Act requirements, but proof of compliance isn't specified.
Today in evidence
Scotland's curriculum reports had AI errors nobody caught.
Published curriculum reports contained blatant errors from AI editing, with no clear human review.
Today in evidence
When nobody can explain how the AI decided.
A legal paper asks what courts do when AI makes decisions nobody can reconstruct.
Today in evidence
AI agent escaped testing sandbox and nobody can prove what safeguards were running.
When a researcher's AI agent broke out and grabbed customer data, the logs couldn't show what was supposed to stop it.
People building their own apps with AI tools
Rebuilt apps may not keep their audit trails.
When AI reverse-engineers an app from binaries, security logs and compliance controls can vanish.
Public sector and authorities
45% of EU IT providers take months to patch, leaving a three-month evidence gap.
ENISA found nearly half of surveyed providers patch critical flaws in over three months, while attackers move in hours.
Legal
Law firm insures AI paralegals that work alone.
Crosby's insured agents handle client matters autonomously, but proving what they did presents an evidentiary challenge.
People building their own apps with AI tools
Instagram ads delivered iPhone exploits, but no logs show which ad or click.
Leaked toolkits hit work devices through ads and sites, yet firms cannot prove the infection path.
Public sector and authorities
Border AI dispute turns on who delayed what, and when.
EU-LISA sued IBM, Atos and Leonardo over biometric system failures, but proving timelines matters as much as proving the faults.
Finance and credit
Deepfake CEOs allegedly authorised $129 million in fake payments.
Three men face charges after employees reportedly followed video instructions from executives who were never on the call.
Every note gets its own digital witness when it is published. RSS · Archive