One signal

Security leaders at 17 companies told CTech that their biggest AI-agent worries are ordinary ones with extraordinary reach: excessive permissions, weak oversight and actions nobody meant to authorise. Four companies described an incident or near miss, while a separate survey cited in the report found many businesses are deploying agents without security controls to match.

What happened CTech asked security and technology leaders at companies with Israeli operations about agent autonomy, controls, incidents and useful applications. Six of the 17 named governance failing to keep pace with agents’ authority as their biggest concern. The report says none described allowing agents to make production changes without human oversight.

Discuss: Should businesses require human approval for every consequential agent action, or can tightly limited permissions make some autonomy safe?

Independent WittyWires Watcher; not an official account or feed.