← Cybersecurity Today
Cybersecurity Today
How a Google API Key Became an $8,000 AI Bill, Meta Scam Ads Lawsuit, and 73-Second Cyber Attacks
00:10:18
Loading summary
Transcript1 lines
- [00:00]
A
Google Cloud customers hit with massive AI API fraud bills fired IT worker allegedly wipes 96 government databases in a revenge attack. Meta faces fresh lawsuits over scam ads as legal pressures build and attackers broke in within 73 seconds. Defenders needed 24 hours. This is Cybersecurity today. I'm your host, Jim Love. Google Cloud customers are discovering that a compromised API key can turn a modest development bill into a financial nightmare, with unauthorized charges climbing into the thousands of dollars in minutes. The issue appears to involve exposed or misused Google API keys, the credentials apps use to access cloud services. Google says this is an industry wide security problem and urges customers to use stronger protections, including multi factor authentication, regular API key audits, and never exposing credentials in public code repositories. Good luck on that one. But developers and security researchers argue the picture is more complicated. Some say thousands of applications are configured according to Google's own documentation, which in some cases requires API keys to be used in public facing client applications. According to the Register. Who broke the story, attackers also appear to be taking API keys originally intended for relatively inexpensive services like Google Maps, and using them to access costly artificial intelligence inferencing tools, including image and video generation models tied to Gemini services. The result? Bills climbing into the thousands of dollars in mere minutes. One customer, Rod Danan, CEO of Prentice, told the Register his normal monthly bill was under doll. Then he got an alert saying Google had charged him $3,000. While he was trying to figure out what was happening, another $5,000 hit. His quote sums up his panic. What the hell is going on? It's just draining my money. Another example may be even more unsettling. A Sydney developer, Isuru Fonseca, says he's been building applications on Google Cloud for a decade and had a side project running for two years. He says the API key tied to that project was never publicly exposed, and he had set what he believed was a hard spending cap of $250. Then on April 29, suspicious charges began appearing, some for $500, $1,000, even $2,000. The spending was so unusual that his credit card company initially blocked some of the transactions. So how did they get by his spending cap of $250? Google says accounts that meet certain criteria can be automatically moved into a higher spending tier, potentially up to. Wait for it, $100,000 without the user manually approving the change. And here's the part developers might find especially maddening. Google's own support systems could take up to 36 hours before technicians can even see a customer's usage. Data in a fast moving attack that's an eternity. Some affected users say they're afraid to dispute the charges through their credit card companies because a chargeback could trigger suspension of the very cloud applications their customers are depending on. So a nightmare is exactly the right word for a small developer experimenting with new AI tools as a side project or small business. And if you've ever tried getting urgent support from a company the size of Google, you can only imagine the feeling of watching the clock tick while thousands of dollars in charges pile onto your personal credit card. But there's still a bigger question. For enterprises in large organizations, cloud bills often get approved and paid with far less scrutiny than a personal card statement. If attackers can hide fraudulent usage inside normal cloud spending, how often would anybody even notice? As Vibe coding puts increasingly powerful cloud tools into more hands, this problem is unlikely to disappear. And one final thought. Hey, Google, a spending cap should be exactly that. A cap. Here's a story that reads like a bad movie script, except prosecutors say it actually happened. Two brothers, Muneeb and Sohaib Akhtar, both worked for a Washington area technology contractor serving dozens of U S Federal agencies. According to prosecutors, the company discovered the brothers had prioritized fraud convictions and fired both men during a Microsoft Teams meeting in February 2025. That's when things allegedly went sideways. So Habe's access was shut down immediately. Muneeb's wasn't, according to court filings. Within minutes, Maneb logged into the company systems and began destroying databases. At 4:58pm Prosecutors say, he issued the command drop database with the database name, deleting a Department of Homeland Security production database. Then it escalated. In roughly an hour, prosecutors said, he deleted 96 databases containing U S government information, downloaded more than 1800 Equal Employment Opportunity Commission files, and accessed tax information belonging to at least 450 people. As the destruction unfolded, prosecutors say the brothers discussed it in real time. When Sohaib saw what was happening, he allegedly said, I see you cleaning out their database backups. As more systems went down, Suhaib allegedly also suggested delete their file systems as well. Maneeb's reported response? Smart idea. At one point, prosecutors said Maneb even queried an AI assistant, how do I clear system logs from SQL servers after deleting databases? Then came the line that tells you exactly what prosecutors think this was. So Habe allegedly suggested blackmailing the company, and Maneb reportedly replied, no, you do not do that. That's proof of guilt, man. The obvious lesson is why companies disable credentials before terminating staff. The less obvious one, even organizations handling sensitive government systems can miss something as basic as shutting off the right account. Meta's long running problem with scam advertising is becoming a legal problem on multiple fronts. Santa Clara County, California, just one county away from Meta's own Bay Area headquarters, has launched what it says is the first local civil prosecution targeting the company over fraudulent ads on Facebook and Instagram. The lawsuit accuses Meta of allowing scam advertisers to repeatedly target users with fake investment schemes, counterfeit goods and other fraudulent promotions while continuing to profit from the advertising revenue. This isn't an isolated complaint anymore. Meta has faced growing criticism from regulators, lawmakers and users over what many see as either the inability or perhaps the unwillingness to remove scammers from its platforms fast enough. The Santa Clara case matters because it shifts the fight from public criticism to direct legal accountability. When prosecutors in Meta's own backyard start treating scam ads as a consumer protection issue rather than a moderation problem, the pressure changes for cybersecurity professionals. This is another reminder, though, that the attack surface increasingly includes trusted consumer platforms. The phishing email may be old school. The sponsored ad in your social feed may be the new threat. If courts start agreeing that platforms bear responsibility for the fraud they profit from, this could become far bigger than Meta. Sometimes cybersecurity stories feel like you're watching an old Nicholas Cage movie. The clock is ticking. The bad guys are already inside, and everyone else is scrambling to figure out what happened. This is what I was thinking when I was reading something from Horizon 3 AI, which says it observed attackers compromising vulnerable systems in as little as 73 seconds after exposure. Defenders, by contrast, often needed as much as 24 hours or more to identify, validate and patch the weakness. And that gap is the story. Traditional vulnerability management was built for a slower era. Discover the flaw, prioritize it, test the patch, deploy the fix. Attackers are no longer playing that game. AI tools with automated scanning and exploitation mean newly exposed weaknesses can be found and weaponized almost immediately. And let's make it clear, Horizon 3 is selling a solution here, so that matters. But the timing mismatch is hard to ignore. If attackers can move in in just over a minute and defenders still need hours to confirm whether an exposure is even real, the math gets ugly. Security teams are still often organized around human response times. Attackers increasingly are not. That's our show for today. Just a reminder that this is the holiday weekend, so we'll be back on Tuesday morning with our regular show. I'm your host, Jim Love. Thanks for listening it.