API Security Is Getting Weird

So yeah… API security just had another moment.

Not the usual “oh someone forgot auth again” type. More like:

“everything was fine… until it wasn’t… and now you owe $100K”

Let’s dive in

1. “Public” API keys are basically not public anymore

You know those API keys devs put in frontend code? Yeah… those can now access AI APIs if enabled.

With the introduction of AI services like Gemini, those same keys can now access powerful APIs if enabled. They’re no longer passive — they can actually perform actions, and those actions cost money.

No exploit. No hack.

Just:

“oh cool free key lemme spam this endpoint real quick”

Unfortunately, a 24-year-old solo developer has already faced this problem — their startup nearly collapsed after an attacker used a publicly accessible key to flood Gemini AI with inference requests. (You can read his story on Reddit)

What changed?

Nothing… and that’s the scary part.

Platforms added new AI capabilities and reused existing API keys for access. That’s it. But suddenly, old keys gained new privileges — without any warning or visibility.

No alerts. No warnings. No “hey btw this is dangerous now”.

oh btw this is called: “retroactive privilege escalation” (fancy name, very cursed concept)

2. New attack meta: just make you broke

Attackers are starting to shift focus. Attackers don’t even care about your data anymore.

Why bother when they can just:

1. Find an exposed API key
2. Hit expensive endpoints (AI, compute, etc.)
3. Let your billing system take the hit

Congrats, now you have a massive bill at the end of the month, waiting for you.

No data breach report needed 👍

3. Meanwhile… classic API issues still exist

While all this AI chaos is happening, we are STILL seeing:

• Exposed admin APIs
• No auth on internal endpoints
• Full RCE because “it’s internal bro”

A recent example is an Apache ActiveMQ vulnerability that allowed remote code execution via its Jolokia API. The issue wasn’t exotic — it was the same old pattern: something internal ended up exposed.

Translation: “we exposed an internal API to the internet and now we’re surprised”

4. AI is becoming an API attack surface

This one is actually insane.

AI systems today don’t just respond — they act. They call APIs, access tools, and execute workflows.

That creates a new kind of attack path. Instead of directly targeting the API, an attacker can manipulate the AI itself:

1. Write a malicious prompt
2. AI reads it
3. AI goes “sure boss”
4. Calls internal APIs
5. Leaks data / does actions

You didn’t get hacked.

Your AI did it for them. The AI was doing exactly what it was designed to do — just not in a way you expected.

The real shift

API attacks used to rely on clear technical flaws like injection or auth bypass. Now, they’re increasingly about abusing systems as designed — just at scale or in unintended ways.

Now it’s more like:

• “let me abuse your billing”
• “let me trick your AI into calling your own APIs”
• “let me use your system as intended… but at scale”

Final thought

The scariest part?

Nobody really “messed up” in these cases.

Devs followed the rules… followed best practices… And ended up exposed.

Which means that The rules are outdated.