惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

Apple Machine Learning Research
Apple Machine Learning Research
J
Java Code Geeks
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
freeCodeCamp Programming Tutorials: Python, JavaScript, Git & More
Last Week in AI
Last Week in AI
雷峰网
雷峰网
博客园_首页
小众软件
小众软件
美团技术团队
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
腾讯CDC
P
Proofpoint News Feed
MongoDB | Blog
MongoDB | Blog
Google DeepMind News
Google DeepMind News
MyScale Blog
MyScale Blog
U
Unit 42
The Cloudflare Blog
钛媒体:引领未来商业与生活新知
钛媒体:引领未来商业与生活新知
Microsoft Security Blog
Microsoft Security Blog
大猫的无限游戏
大猫的无限游戏
Engineering at Meta
Engineering at Meta
N
Netflix TechBlog - Medium
Microsoft Azure Blog
Microsoft Azure Blog
博客园 - 叶小钗

Consumer Insights

The ransomware negotiator who was working for the other side After years on the run, alleged Ryuk ransomware operator pleads guilty INTERPOL crackdown shows scammers shifting to social media Meta lets strangers remix your public Instagram photos with AI—here’s how to opt out Invited to a "job interview" with Netflix or OpenAI? Beware! Your Google password could be at risk Two arrested over credit card phishing - as the Netherlands is named Europe's worst for payment fraud India pauses WhatsApp username feature over security concerns Alleged teen ransomware hustler faces US charges after arrest in Finland WhatsApp usernames explained: how to reserve yours and stay safe Scammers race to cash in on Venezuelan earthquake disaster USB drives carrying China-linked malware infected Japanese military networks for nearly a year WhatsApp tests new safety prompt before you chat with strangers Social media is worth celebrating. It's also worth protecting. Polish police dismantle SIM-swap gang accused of crypto theft Operation Endgame deals fresh blow to StealC and Amadey malware networks Hacker hijacks Brazil's national alert system, sending "misanthropy" to millions of phones Cybercrime now rivals traditional crime across parts of Asia Apple's Hide My Email tweak leaves privacy fans fuming Americans lost $3.5 billion to imposter scams last year — and the scams are getting harder to spot Scammers have killed the physical Steam Gift cards Crypto investment scam sends couriers to collect victims' cash, FBI warns Maine forced to take down data breach portal after fake notices filed with authorities Privacy own-goal: World Cup blunder leaks Lionel Messi's passport details Why schools remain one of cybercriminals' favourite targets WhatsApp detects new spyware activity from Israel’s NSO Group despite court order Got a LinkedIn message from a recruiter? It might be Chinese intelligence, warn FBI and MI5 Europol cracks down on illegal streaming globally Hackers didn't hack Instagram; they just asked Meta AI FBI Warns Fans About FIFA Scams Ahead of 2026 World Cup Virtual knife, real lawsuit: Counter-Strike skin dispute ends in court
Deleted Google API keys may remain active for 23 minutes
Vlad CONSTANTINESCU · 2026-05-22 · via Consumer Insights

Google API keys may keep working after deletion, giving attackers time to abuse Gemini access and generate costly cloud charges.

A risky revocation gap

Security researchers at Aikido say deleted Google API keys can continue authenticating after users remove them, with the longest observed window reaching 23 minutes. As many developers treat deletion as an emergency brake after sensitive data leaks, this finding demonstrates how simply removing impacted components isn’t always a silver bullet.

Aikido ran 10 trials, sending repeated authentication requests after deleting newly created keys. Some Google systems rejected the credentials quickly, but others kept accepting them until revocation propagated.

Gemini access raises the stakes

The risk extends beyond nuisance traffic, as the research shows a stolen key tied to a project with Gemini enabled could be exploited to access uploaded files, cached context and billable AI services before the key finally stops working.

That delay grows even more dangerous when threat actors automate high-value requests. Even if jus some backend servers still honor the deleted credential, repeated attempts increase the odds of successful abuse.

Billing controls may not stop the damage

The discovery comes amid growing concern over runaway AI bills. Google’s Gemini API billing model employs usage tiers based on payment history, and Google says higher tiers can reach spending caps from $20,000 to $100,000 or more.

Google recently introduced project spend caps and dashboards, but the research suggests response plans should not rely on key deletion alone. Developers should restrict keys by service and client, avoid exposing them in code or URLs, monitor unusual usage and rotate credentials proactively.

Google reportedly declines a fix

Aikido says Google classified the behavior as intended and closed the report as infeasible to fix. The researchers argue the delay is technically solvable, citing faster propagation for other Google credential types and newer Gemini key formats.

Cloud and AI developers should treat leaked Google API keys as active even after deletion. Incident response should include disabling affected services, reviewing activity, checking cached data and contacting billing support immediately.