Opus 4.7 Disappoints: Paranoid Safety Checks Slow Down an Otherwise Unchanged AI Model

Opus 4.7 is out and it is underwhelming. Also it looks like they cranked up the "malware" detection to the max.

I spent hours yesterday putting it through its paces, and honestly? It felt like working with 4.6 all over again.

The performance seems nearly identical. Same reasoning quality, same creative output, same occasional quirks we've grown used to.

But here's the weird part - every few seconds in its thought process, it would pause to ask itself: "Am I accidentally helping create malware here?" Over and over again. Even for the most innocent product features.

It's like having a paranoid assistant who constantly second-guesses whether your marketing copy might somehow be dangerous. Slows things down and breaks the flow.

My guess? They're probably stress-testing their safety systems before rolling out Mythos. Makes sense from a risk management perspective, but man, it makes the current experience clunky.

Are you seeing similar over-cautious behavior in your AI tools, or is this just an Anthropic thing?