As if email senders don’t have enough to worry about, here’s one more headache: agentic agents.

These robotic agents can infiltrate email systems, and just about anything else, and they can harm companies just as  severely as traditional hackers, judging by a study from Ironscales. 

Of those polled, 77% say an AI agent acted on malicious or misleading content over the past 12 months. And 57% experienced a confirmed attack.

The problem could be spreading, because 96% of companies are using AI agents for chores that would normally be performed by employees. And 80% are allowing agents into their internal systems. 

Perhaps worse, 67% of companies say agents can reach their ticketing and helpdesk tools, 56% cloud storage, 49% internal file repositories and 47% securities tools and logs. 

"For years, the goal of a phishing email was to fool a person,” says Eyal Benishti, CEO of Ironscales. “Today it only has to fool the agent sitting between that person and their inbox.”

Benishti adds, “Organizations have handed AI agents the keys to their email, files, and even payment systems, and attackers are using the oldest tricks in the book to turn those agents against them.”

Rest assured that cyber felons are on top of this opportunity.  Of the firms that suffered an AI agent-related attack, the most common types were an AI tool acting on a malicious email or message (42%) or a phishing or social engineering attempt (41%).

Who is responsible for letting these attackers in? IT is the main  approver of AI agents at 78% of companies. Security, which you think would have this authority, fulfills this role at only 6%.

Working with Censuswide, Ironscales surveyed 501 U.S.-based IT and cybersecurity professionals at the middle manager level and above at firms with from 1,000 to 10,000 employees in September.

On another front, 54% of the firms that use agents report that one had made a mistake in the past 90 days, according to a study from Nylas. Most often, this happened when a workflow was triggered that had to be reversed. And only 53% of users can produce a full audit log of what their agents did in the past week. 

Over 50% of the respondents are now running six or more agents. Of those that run one to five, 45% have reported an incident, versus 70% who run more than 20. 

"The category has moved from 'will this work' to 'what do we do when it doesn't,'" said Jeff Koets, CEO of Nylas. "More than half the companies in this survey have already answered that question at least once, by reversing a workflow, correcting a record, or apologizing to a customer.”

Nylas surveyed 1,030 U.S.-based professionals in August and September through the Centiment research panel.