CYBERSECURITY

Morning Brew Inc, Photos: Warner Brothers

Early on in the 1982 sci-fi thriller Blade Runner, agent Rick Deckard (Harrison Ford) puts a bioengineered, human-like being known as a “replicant” to the test—specifically, a “Voight-Kampff” test. The tense exam measures interviewee responses to a series of odd or emotionally provocative questions and statements.

For example, Deckard asks: “You’re watching television. Suddenly you notice a wasp crawling on your arm.”

Careful with your answer. In the movie, your answer determines if you’re human, which means Deckard will let you live.

While today’s cybersecurity pros don’t have time to sit across from every bot and ask it ethical questions, there are plenty of real-life, modern security mechanisms (like a CAPTCHA) to determine a human visitor from a fake one.

Like Deckard facing a new generation of even more human-y replicants, the job of sorting bots from humans has only gotten tougher thanks to the rise of AI agents and sophisticated deepfakes. That job applicant on a video call, for example, could be a digital fake using an AI-generated knowledge base to answer questions.

A cybersecurity pro separates fact from science fiction.—BH

      

Sponsored By SentinelOne

Attackers have leveled up. Today’s threats move at the speed of AI with enterprise-grade sophistication—and you’re at risk if your EDR stack can’t keep up.

Tune in as SentinelOne experts Chris Hosking and Anthony La Scala debunk five EDR myths in The Endpoint Stress Test webinar. They’re pressure-testing common misconceptions in endpoint security against what growing organizations actually need.

Save your spot to learn about:

• what modern detection catches that legacy EDR wasn’t built for
• the risks of treating identity as a separate tool in a modern attack surface
• how a unified data fabric cuts through alert fatigue for understaffed teams

Modernizing your endpoint infrastructure doesn’t mean increasing headcount. Learn more.

CYBERSECURITY

Francis Scialabba

You can’t judge a person unless you’ve walked a mile in their shoes. But you can judge a hacker once you’ve accessed their operational environment.

That was the case for Palo Alto Networks subsidiary Unit 42 when an autonomous agent exposed the infrastructure of an AI-enabled attack sequence by accidentally starting a file server in its home directory.

“We were able to read through their configuration files and their LLM chat logs, for lack of a better word, their instructions they were passing to the LLMs,” Andy Piazza, senior director of threat research at Unit 42, told IT Brew. “So, we could see them testing and trying various attacks.”

The TL;DR on the attack. Unit 42, which detailed its findings in a July 30 research report, said the campaign was executed by a Chinese-speaking threat actor that goes by the aliases “knaithe” and “KnYuan.” The primary tool of choice for the campaign was AI model DeepSeek via the Hermes Agent open-source framework. The Hermes Agent fulfilled the orchestration function, while DeepSeek acted as a reasoning agent (i.e., the agent in charge of decision-making).

DeepSeek wasn’t the only AI tool used in the campaign.—BM

      

WORD OF THE WEEK

Observability tools: Observability, or a clear view into a system or network, is one of the most crucial functions of any IT team, especially as tech stacks become more complicated. Once a team has a holistic view of how something is operating, they can quickly diagnose and solve issues as they arise. Learn more.

CLOUD

Getty Images

Are you keeping an AI on your agent?

That’s the kind of observability question required at a time when the tech industry is embracing the capabilities of AI tools and giving agentic AI more autonomy as solutions grow in complexity. For Michael Bevilacqua, VP of AI product for ETL service provider Adeptia, the question is about data oversight—a focus on telemetry that relies on the data observation being run “headless,” or without human decision-making.

“For CTOs in today’s world, there’s headless observability into the system, citations for every AI decision,” Bevilacqua said. “Those are all paths that are now required in an enterprise product.”

An approach that prioritizes that kind of headless decision-making allows for smart testing, as well, Bevilacqua added, providing full telemetric observability into a company’s ecosystem.

“What they want is a natural language interface into all the different systems, and that’s what I would be demanding from any vendor that brought products into my door,” Bevilacqua said.

Where does accountability lie?—EH

      

patch notes

Francis Scialabba

Today’s top IT reads.

Stat: 10%. That’s how much of the English-language internet—or a big snapshot of it, at least—shows “significant signs of AI authorship.” (Pew Research Center)

Quote: “We have every reason to believe that some unscrupulous employers are taking the robot tips for themselves.”—Holona Ochs, an associate professor of political science at Lehigh University, on tips given to robots working in cafés and other venues (BBC)

Read: Thirteen tools that test how well your agent performs. (CIO)

^{*A message from our sponsor.}

Level up your career with these resources from our sponsors!