What's this about?
Known Agents, which measures AI bot traffic across 5,000+ websites, disclosed an active campaign in which an attacker impersonates well-known AI crawlers like ClaudeBot to scan websites for vulnerabilities. What separates this from ordinary background scanning is what it's looking for: the most frequently targeted path is `/.config/anthropic/credentials/default.json` — the attacker is hunting for the credential and configuration files that AI coding tools leave on disk. The same report shows bots are now 35% of all web traffic, with the AI-related share of that jumping 11 points to 29%.
What exactly is "spoofing a bot"?
Every program that visits a website announces who it is in a text field called the user-agent — "I am ClaudeBot," "I am Googlebot." Nothing about that field is verified by default; it's a name tag you write yourself. Legitimate crawlers therefore publish a list of IP addresses they crawl from, or cryptographically sign their requests (Web Bot Auth), so a site can check the claim. A spoofed visit is one that wears the name tag but fails that check — and attackers wear AI-bot name tags because many sites now allow them through.
What's the community saying?
- "Water is wet." The most common first reaction was that mass scanning is the permanent background state of the internet — every server with port 443 open gets thousands of probes daily looking for WordPress logins. The new part is only the disguise.
- Blocking by network, not by name. The most practical thread: since user-agents are trivially faked, look up which ASN owns the IP instead. Several people reported that blocking most VPS providers makes the majority of fake bots vanish — though residential proxies and hijacked mobile apps still get through.
- Why pretend to be an AI bot at all? Genuine puzzlement here, since AI crawlers are increasingly *blocked*. The best hypotheses offered were reputational sabotage of AI companies, or simply that enough sites now allowlist AI bots to make it profitable.
- Defense in depth beyond Cloudflare. Several commenters argued for a second, non-Cloudflare layer behind the CDN, on the theory that a single vendor becomes a single point of failure when a bypass is discovered.
- "Is the open internet over?" The philosophical tail of the thread: a well-received comment suggested this pushes the web toward mandatory identity verification for all traffic, recreating a Facebook-style KYC world at the protocol level.