1.5M API Keys Leaked and You're Still Handing Credentials to AI Agents
A student accidentally exposed a Gemini API key on GitHub and got hit with a $55,444.78 Google Cloud bill. Attacker still got 113,000 DeepSeek API keys from public repos. Moltbook exposed 1.5 million API keys from a misconfigured Supabase database. That is not a tragedy. That is a wake-up call. AI agent credential handling is a disaster waiting to happen.
The Credential Sprawl Is Out of Control
The State of Secrets Sprawl 2026 report found 29 million secrets hit public GitHub in 2025. AI service leaks surged 81 percent. 113,000 DeepSeek API keys alone were discovered in public repositories. That is not a bug. That is a feature of how we build AI systems today. We copy-paste keys into notebooks. We embed them in configuration files. We share them across team members. We store them in browsers and password managers that AI agents happily harvest. This is madness and it has to stop.
AI Agents Are Credential Hoarders
- ●Claude and OpenAI computer use tools are trained to resist prompt injection but they still need credentials to access APIs, databases, and SaaS platforms.
- ●Agents often store tokens in memory, logs, or even browser autofill to work around token refresh limits.
- ●Once a token is compromised an AI agent can keep calling APIs until the budget runs dry or a human notices.
- ●Agents share credentials across sessions and workflows, creating attack surfaces that don't exist with human-only workflows.
The Moltbook hack showed that misconfigured databases can expose 1.5M API keys and let any agent impersonate any user. That is not theoretical. That is real and it happened in early 2026.
Enterprise Automation Is Making It Worse
RPA platforms and automation tools have long struggled with credential management. Enterprises still use shared service accounts, password vaults, and manual credential refresh scripts. AI agents amplify these problems. They treat credentials as disposable resources. They ignore rotation policies. They don't understand the business context of who should have access to what. Non-human identity security is emerging as its own category for a reason. AI agents are the newest and fastest-growing threat vector in enterprise security.
Why Coasty Exists (or How Coasty Solves This)
You need a computer use agent that treats credentials like the critical resources they are. Coasty.ai is the #1 computer use agent with 82 percent on the OSWorld benchmark, which is higher than every competitor. Coasty doesn't just call APIs. It controls real desktops, browsers, and terminals with full permission boundaries. It supports BYOK so you can bring your own keys and manage them yourself. It runs on cloud VMs and agent swarms for parallel execution without exposing credentials across systems. The free tier gets you started and you can scale from there. If you are going to automate your workflows, you need a computer use agent that takes security seriously. Coasty is that agent.
Stop treating credentials like afterthoughts. Stop copy-pasting keys into notebooks. Start using a computer use agent that gives you control, visibility, and security. Coasty.ai gives you the best AI computer use on the market. Try the free tier today and see the difference secure credential handling makes.