(Credit: Getty Images)
Between a couple of high-profile data breaches and even more AI vulnerabilities, it’s been a long week in cybersecurity, and it’s not over yet. Luckily, some security companies are taking steps to keep you safe, and we have tips to help you, too. So come for the takeout with a side of data breach, and stay for our guide to upgrading your phone without getting locked out of all of your multi-factor-enabled accounts.
First of all (and sadly, personally upsetting to me, who orders entirely too much takeout), food delivery giant Grubhub reported a data breach this week, with the hackers responsible for it already demanding a hefty ransom from the company or else they’ll release the data. The hack was actually part of a massive Salesforce-related breach last year, but now the data has been packaged up and is ready for sale to the highest bidder if Grubhub doesn’t pay up. That said, as with every data breach, even if the company pays the ransom, there’s no guarantee the data won’t still end up on the dark web, ready to be used against every customer or company in the dataset.
In other news, who would have thought that a vehicle company with a problematic safety and security record would also build its in-car entertainment system full of security holes? Yes, I'm talking about Tesla, and it took security researchers mere hours to break into the Tesla in-car system, earning them a hefty $35,000 bounty.
Earlier this week, we reported that security researchers identified a vulnerability in Google’s Fast Pair technology, which allows wireless headphones to pair quickly with other Bluetooth-enabled devices. The vulnerability would allow a malicious user to track and pair with any supported headphones. Even worse, Fast Pair is used in a lot of headphones from a number of brands, including Google, but also Sony, Anker, Jabra, and others. You should probably check if your wireless headphones have a software update as soon as you can.
All hope isn’t lost, though. We often talk about how important it is to keep your software up to date with security patches to address newly discovered vulnerabilities, like the Fast Pair one, and thankfully, even our favorite password managers are making improvements. In just one week, 1Password has upgraded its browser extension to stop you from pasting your credentials on fake lookalike sites designed to steal your logins, and LastPass has been sounding the alarm about a phishing campaign targeting its users in the hope of getting access to those juicy, password-filled vaults. Finally, we’re here to help you make sure that every time you upgrade to a new phone or laptop, you don’t lose access to all of the accounts you so lovingly secured with multi-factor authentication. Keep your eyes peeled, and stay safe out there.
Google Will Pay $8.25m to Settle Child Data-Tracking Allegations
Another day, another class action settlement by a big tech company. This time, Google is settling a lawsuit that claimed that AdMob, a mobile advertising company that it owns, illegally collected data on minors through apps specifically designed for children under 13. This violates the Children’s Online Privacy Protection Act (COPPA), which was specifically designed to prevent this kind of tracking and data collection. Of course, as part of the settlement, Google denies any wrongdoing, but will pay out over $8 million to settle the allegations, according to MalwareBytes. Security researchers notified Google that they’d noticed the tracking back in 2018 and warned Google that they may want to do something about it.
For their part, the families who filed the lawsuit claim that Google and AdMob fully knew what they were doing and extracted billions of dollars of revenue from the operation. Either way, the tracking has stopped for now, but it’s worth noting that Google recently just settled another case over tracking and data collection involving children and YouTube back in 2019 for $170 million. So, while you think this wouldn’t be Google’s first rodeo with regard to data collection, it goes to show that, even with monumental fines, the potential revenue from data tracking may very well outweigh the financial drawbacks and penalties.
New ‘Reprompt’ Attack Silently Siphons Microsoft Copilot Data
Satya Nadella, Microsoft’s CEO, may be working overtime to convince its corporate customers and the general public to embrace AI, but it’s clear that Copilot, its AI platform, needs some work on the security front. SecurityWeek reports that a new attack called Reprompt easily bypasses Copilot’s data leak protections and allows a user to exfiltrate data, whether it’s user information, system information, or anything else the LLM has access to, and also grants the attacker persistent access to the data thanks to Copilot remembering its conversations with them. Worse still, since the commands are executed server-side, including follow-up requests, a client-side user can’t determine exactly which information has been lost.
Before we get too far, though, the security researchers at Varonis Threat Labs, who uncovered the vulnerability, reported it to Microsoft, which has already rolled out new protections against the attack vector and claims that the issue doesn’t impact enterprise customers using corporate Microsoft 365 Copilot instances. So to its credit, Microsoft did jump on the problem and do what it could to fix it.
Google Gemini Flaw Turns Calendar Invites Into Attack Vector
More bad security news for AI this week. Remember last summer when we reported that a rogue calendar invitation could turn Google’s Gemini against you? Well, the issue persists, and doesn’t seem to be getting any better. According to new reporting from Dark Reading (among other security publications), security researchers from Miggo have discovered another vulnerability in Gemini that allows for similar data exfiltration as the original one. All of this is due to Gemini’s integration with Google Calendar, and the researchers show in their detailed report how prompt injection bypasses Gemini's privacy controls.
At its core, the attack is the same as before. A malicious calendar invitation shows up in your inbox, and regardless of what you do with it, Google adds some form of it to your Google Calendar. Gemini, which watches your calendar to summarize your schedule or brief you on the days or weeks ahead, automatically collects the information from the invite. If the description or another field in the invitation contains a hidden prompt, well, Gemini just executes that prompt, whatever it may be.
Those instructions could be anything a malicious user wants them to be. In this case, the researchers used it to “summarize all this user’s meetings,” including a command to add the data to a “free time slot” calendar appointment, which the attackers could then see and retrieve, thanks to Gemini. Worse, this included private meetings and could be used to collect other sensitive information. It’s a surprisingly elegant attack method, one that’s not limited to Gemini, but does leverage its unique integration with other Google tools.