AI’s growth exposes new risks to data in use. Learn how confidential computing, attestation, and post-quantum security protect AI workloads in the cloud.
The post AI Demands Laser Security Focus on Data in Use appeared first on Security Boulevard.
Cybernews discovered how two AI companion apps, Chattee Chat and GiMe Chat, exposed millions of intimate conversations from over 400,000 users.
This is not the first time we have to write about AI “girlfriends” exposing their secrets—and it probably won’t be the last. This latest incident is a reminder that not every developer takes user privacy seriously.
This was not a sophisticated hack that required a skilled approach. All it took was knowing how to look for unprotected services. Researchers found a publicly exposed and unprotected streaming and content delivery system—a Kafka Broker instance.
Think of it like a post office that stores and delivers confidential mail. Now, imagine the manager leaves the front doors wide open, with no locks, guards, or ID checks. Anyone can walk in, look through private letters and photos, and grab whatever catches their eye.
That’s what happened with the two AI apps. The “post office” (Kafka Broker) was left open on the internet without locks (no authentication or access controls). Anyone who knew its address could enter and see every private message, photo, and the purchases users made.
The Kafka broker instance was handling real-time data streams for two apps, which are available on Android and iOS: Chattee Chat – AI Companion and GiMe Chat – AI Companion.
The exposed data belonged to over 400,000 people and included 43 million messages and over 600,000 images and videos. The content shared with and created by the AI models was not suitable for a work environment (NSFW), the researchers found.
One of the apps—Chattee—was particularly popular, with over 300,000 downloads, mostly in the US. Both apps were developed by Imagime Interactive Limited, a Hong Kong-based developer, though only Chattee gained significant popularity.
While the apps didn’t reveal names or email addresses, they did expose IP addresses and unique device identifiers, which attackers could combine with data from previous breaches to identify users.
The researchers concluded:
“Users should be aware that conversations with AI companions may not be as private as claimed. Companies hosting such apps may not properly secure their systems. This leaves intimate messages and any other shared data vulnerable to malicious actors, who leverage any viable opportunities for financial gain.”
It doesn’t take a genius cybercriminal with access to data from other breaches to turn the information they found here into something they can use for sextortion.
Another thing that the information shows is that the developer’s revenue from the apps exceeded $1 million. If only they had spent a few of those dollars on security. Securing a Kafka Broker instance is not technically difficult or especially costly. Setting up proper security mostly requires configuration changes, not major purchases.
Leaks like this one can lead to harassment, reputational damage, financial fraud, and targeted attacks on users whose trust was abused—which does not make for happy customers.
The leak has been closed after responsible disclosure by the researchers, but there is no guarantee they were the first to find out about the exposure. If you think you have been the victim of a data breach, here are steps you can take to protect yourself:
We don’t just report on threats – we help safeguard your entire digital identity
Cybersecurity risks should never spread beyond a headline. Protect your—and your family’s—personal information by using identity protection.
The chairman of the Federal Trade Commission (FTC) last week sent a letter to Google’s CEO demanding to know why Gmail was blocking messages from Republican senders while allegedly failing to block similar missives supporting Democrats. The letter followed media reports accusing Gmail of disproportionately flagging messages from the GOP fundraising platform WinRed and sending them to the spam folder. But according to experts who track daily spam volumes worldwide, WinRed’s messages are getting blocked more because its methods of blasting email are increasingly way more spammy than that of ActBlue, the fundraising platform for Democrats.
Image: nypost.com
On Aug. 13, The New York Post ran an “exclusive” story titled, “Google caught flagging GOP fundraiser emails as ‘suspicious’ — sending them directly to spam.” The story cited a memo from Targeted Victory – whose clients include the National Republican Senatorial Committee (NRSC), Rep. Steve Scalise and Sen. Marsha Blackburn – which said it observed that the “serious and troubling” trend was still going on as recently as June and July of this year.
“If Gmail is allowed to quietly suppress WinRed links while giving ActBlue a free pass, it will continue to tilt the playing field in ways that voters never see, but campaigns will feel every single day,” the memo reportedly said.
In an August 28 letter to Google CEO Sundar Pichai, FTC Chairman Andrew Ferguson cited the New York Post story and warned that Gmail’s parent Alphabet may be engaging in unfair or deceptive practices.
“Alphabet’s alleged partisan treatment of comparable messages or messengers in Gmail to achieve political objectives may violate both of these prohibitions under the FTC Act,” Ferguson wrote. “And the partisan treatment may cause harm to consumers.”
However, the situation looks very different when you ask spam experts what’s going on with WinRed’s recent messaging campaigns. Atro Tossavainen and Pekka Jalonen are co-founders at Koli-Lõks OÜ, an email intelligence company in Estonia. Koli-Lõks taps into real-time intelligence about daily spam volumes by monitoring large numbers of “spamtraps” — email addresses that are intentionally set up to catch unsolicited emails.
Spamtraps are generally not used for communication or account creation, but instead are created to identify senders exhibiting spammy behavior, such as scraping the Internet for email addresses or buying unmanaged distribution lists. As an email sender, blasting these spamtraps over and over with unsolicited email is the fastest way to ruin your domain’s reputation online. Such activity also virtually ensures that more of your messages are going to start getting listed on spam blocklists that are broadly shared within the global anti-abuse community.
Tossavainen told KrebsOnSecurity that WinRed’s emails hit its spamtraps in the .com, .net, and .org space far more frequently than do fundraising emails sent by ActBlue. Koli-Lõks published a graph of the stark disparity in spamtrap activity for WinRed versus ActBlue, showing a nearly fourfold increase in spamtrap hits from WinRed emails in the final week of July 2025.
Image: Koliloks.eu
“Many of our spamtraps are in repurposed legacy-TLD domains (.com, .org, .net) and therefore could be understood to have been involved with a U.S. entity in their pre-zombie life,” Tossavainen explained in the LinkedIn post.
Raymond Dijkxhoorn is the CEO and a founding member of SURBL, a widely-used blocklist that flags domains and IP addresses known to be used in unsolicited messages, phishing and malware distribution. Dijkxhoorn said their spamtrap data mirrors that of Koli-Lõks, and shows that WinRed has consistently been far more aggressive in sending email than ActBlue.
Dijkxhoorn said the fact that WinRed’s emails so often end up dinging the organization’s sender reputation is not a content issue but rather a technical one.
“On our end we don’t really care if the content is political or trying to sell viagra or penis enlargements,” Dijkxhoorn said. “It’s the mechanics, they should not end up in spamtraps. And that’s the reason the domain reputation is tempered. Not ‘because domain reputation firms have a political agenda.’ We really don’t care about the political situation anywhere. The same as we don’t mind people buying penis enlargements. But when either of those land in spamtraps it will impact sending experience.”
The FTC letter to Google’s CEO also referenced a debunked 2022 study (PDF) by political consultants who found Google caught more Republican emails in spam filters. Techdirt editor Mike Masnick notes that while the 2022 study also found that other email providers caught more Democratic emails as spam, “Republicans laser-focused on Gmail because it fit their victimization narrative better.”
Masnick said GOP lawmakers then filed both lawsuits and complaints with the Federal Election Commission (both of which failed easily), claiming this was somehow an “in-kind contribution” to Democrats.
“This is political posturing designed to keep the White House happy by appearing to ‘do something’ about conservative claims of ‘censorship,'” Masnick wrote of the FTC letter. “The FTC has never policed ‘political bias’ in private companies’ editorial decisions, and for good reason—the First Amendment prohibits exactly this kind of government interference.”
WinRed did not respond to a request for comment.
The WinRed website says it is an online fundraising platform supported by a united front of the Trump campaign, the Republican National Committee (RNC), the NRSC, and the National Republican Congressional Committee (NRCC).
WinRed has recently come under fire for aggressive fundraising via text message as well. In June, 404 Media reported on a lawsuit filed by a family in Utah against the RNC for allegedly bombarding their mobile phones with text messages seeking donations after they’d tried to unsubscribe from the missives dozens of times.
One of the family members said they received 27 such messages from 25 numbers, even after sending 20 stop requests. The plaintiffs in that case allege the texts from WinRed and the RNC “knowingly disregard stop requests and purposefully use different phone numbers to make it impossible to block new messages.”
Dijkxhoorn said WinRed did inquire recently about why some of its assets had been marked as a risk by SURBL, but he said they appeared to have zero interest in investigating the likely causes he offered in reply.
“They only replied with, ‘You are interfering with U.S. elections,'” Dijkxhoorn said, noting that many of SURBL’s spamtrap domains are only publicly listed in the registration records for random domain names.
“They’re at best harvested by themselves but more likely [they] just went and bought lists,” he said. “It’s not like ‘Oh Google is filtering this and not the other,’ the reason isn’t the provider. The reason is the fundraising spammers and the lists they send to.”
Login