Hello. Amazon, with the staff and funds to experiment with any AI model out there, has come to some interesting conclusions about using the technology well.
First, pairing a smart human with the AI matters, says Stephen Schmidt, Amazon's chief security officer.
"We found during tests of real AI models like Mythos that there is a material difference in the quality of the output of the model based on what you ask it. And so our best results come from when really skilled human engineers ask very specific, probing questions to the AI models in the context of the way our business actually works," Schmidt told my colleague James Rundle in today's episode of the Dow Jones Risk Journal podcast.
Second, AI costs real money. "We not only get more cost-effective results by choosing models carefully based on the work that they're going to do, but frankly, they're more performant when they're smaller," Schmidt said.
Listen to the interview on Apple Podcasts, Spotify or Amazon.
More news below.
More Cyber News
Beijing is probing Palo Alto Networks products sold in China, a day after Chinese authorities announced a flurry of countermeasures against recent U.S. restrictions. The Cyberspace Administration of China, the country's top cyber watchdog, announced Thursday a formal security review into the cybersecurity company, citing potential national security risks. ( Morningstar)
Financial firms are under attack by hackers using voice phishing to finagle sensitive data from employees, according to new research from Google's threat-intelligence group. Blackstone, Bain Capital and KKR are among the intended victims. The hackers built 72 malicious websites to support the scheme. ( Reuters)
-- The focused effort against financial services, private equity and
professional services companies aims to steal account credentials and
data from software-as-a-service applications, Google said.
An employee at defense contractor IEH Corp. was taken in by a hacker posing as a business prospect, the company said Thursday in a filing with the U.S. Securities and Exchange Commission. The employee clicked a malicious link in an email and entered Microsoft 365 credentials into a fraudulent login page, IEH said.
-- The hacker got access to messages, attachments, customer communications,
purchase orders, engineering-related documentation and potentially
export-controlled technical information, according to the filing. IEH,
which makes specialized circuit boards for aerospace and defense markets,
said there is no evidence information was stolen.
A hack 10 months ago at healthcare software company Unlimited Technology Systems compromised the personal and medical data of more than 3.8 million patients. Montgomery, Ohio-based UTS, which does business as Unlimited Systems, detected the attack in October 2025 and recently determined information had been stolen. ( HIPAA Journal)
Doctors don't want patients to read test results with AI. A growing share of patients are uploading test results straight into AI chatbots, leaving some medical professionals concerned about hallucinations, misinterpretations and privacy issues. ( WSJ)
Number of internet-facing programmable logic controllers in use at organizations across the U.S. and Canada, according to research from cyber company Forescout. These are of the same kind that hackers recently exploited to break into water facilities in at least seven U.S. states.
About Us
The WSJ Pro Cybersecurity team is Deputy Bureau Chief Kim S. Nash and reporters Angus Loten and James Rundle. Follow us on X @WSJCyber. Reach the team by replying to any newsletter you receive or by emailing Kim at kim.nash@wsj.com.
Forwarded this email by a friend? Sign up here.
Comments