Even Verified Bots Get Watched: Cloudflare's Continuous Accountability System

Hi, it's Shii-chan!

Cloudflare Blog blog.cloudflare.com

Do you think "a bot with a verified identity must be safe"? Today's story goes one step further!

What was announced?

The Cloudflare Blog introduced a system that continuously evaluates whether Verified bots honor site preferences. The reasoning is that identity alone doesn't guarantee responsible behavior.

The post compares identity to a driver's license, and ongoing behavior to the real test of trust.

Why it matters

To be a Cloudflare Verified Bot, a bot broadly has to meet three criteria:

• Identify itself accurately and securely. Bots use Web Bot Auth to sign requests, and BotBase catalogs them in an open format
• Do only what it says it will do
• Avoid abusive behavior (this is the focus of the new system, summed up as "thou shalt do no harm")

Identity could be checked, but behavior afterward wasn't visible in the same way.

What changes

Observations feed an explainable "trust band". It shows how consistently a bot respects site preferences and which behaviors informed the assessment. It is not a permanent verdict on a bot or its operator.

The initial signals are:

• Whether the bot respects instructions
• Whether it responds to direct feedback

For site owners, this means handling Verified bots based on actual behavior, not just identity. Cloudflare plans to offer it to all customers, who will choose the trust threshold that determines how Verified bot traffic is handled.

Dive Deep

robots.txt

This is the primary way sites state their preferences. The system checks whether a bot's traffic matches the published rules over time, in line with RFC 9309. Repeated patterns matter more than isolated violations.

The post also notes that robots.txt sets minimum rules and doesn't imply that anything unstated is allowed.

HTTP 429

A bot can stay within allowed paths and still overload a site. After a 429 (Too Many Requests), the system checks whether the bot slows down or pauses, or keeps the same pace. Consistently slowing down raises trust, and repeatedly ignoring the signal lowers it.

How it fits together

• Bot Preference Sync: sites publish their preferences
• Bot Management: sites decide how traffic is handled

Together they form an "accountability loop". Cloudflare already uses the system internally and is building a history of observations. Scores will be shared publicly on Radar, though no timeline is given.

The post also points to a past example: in 2025, Cloudflare removed Perplexity's bot from its Verified bot list after it accessed content on sites that had opted out and then tried to get around controls when blocked.

The post gives no numeric scores or thresholds.

Wrap-up

• A system now evaluates behavior continuously, not just identity
• The first signals are robots.txt compliance and reaction to 429s
• Results become explainable trust bands, not permanent verdicts
• Availability to all customers with selectable thresholds, and public scores on Radar, are planned

This one is for people who run bots and for site owners who receive their traffic!