Sacramento Bee's AI Earthquake and Wildfire Bots Spread Misinformation

The Sacramento Bee is learning an uncomfortable lesson about the limits of automated journalism, and Northern California readers are paying the cost in misinformation.

The McClatchy-owned paper has been running two AI-powered bots, the CA Earthquake Bot and the CA Wildfire Bot, to generate real-time coverage of seismic activity and fire incidents across the state. The bots publish dozens of posts per day, pulling from data feeds including USGS earthquake notifications, Cal Fire reports, and National Interagency Fire Center data. The volume is impressive. The accuracy, lately, is not.

On Monday evening, the Earthquake Bot published multiple reports about earthquakes that never happened. One post described a magnitude 3.7 offshore quake west of Bonny Doon in Santa Cruz County. Another reported a 3.2 magnitude event near Shaver Lake. Both were false alarms that originated with the USGS’s own automated notification system, which registered the erroneous events before the agency caught the errors and pulled the notices from its website.

The USGS corrected its records. The Sacramento Bee did not follow suit, at least not quickly. As of this writing, those earthquake posts remain live, now linking out to USGS pages that no longer exist because the agency deleted them after acknowledging the errors.

The wildfire situation produced a more pointed correction. The CA Wildfire Bot published a report about a fire in Ventura County that turned out to be entirely wrong. The bot appears to have confused the United Fire, a 500-acre blaze that broke out Wednesday in Valencia County, New Mexico, with a fire in Southern California’s Ventura County. The two place names share a few letters and nothing else.

Cal Fire noticed the erroneous report and alerted the Bee directly. The paper removed the post and published a correction that reads, in part: “No such fire existed. The report was generated due to a system error from data provided by the National Interagency Fire Center. We apologize for the error.”

That apology matters, and so does the underlying problem it points to. During wildfire season, when evacuations hinge on accurate, real-time information, a false report about a Ventura County fire is not a minor glitch. People make decisions about whether to leave their homes based on exactly this kind of coverage. Automation that produces false positives at scale is not just embarrassing. It can cause real harm.

The Bee’s approach to disclosing its bot-generated content has also shifted in a telling direction. The original disclaimer on these posts had a breezy, self-deprecating tone, noting that “no human journalist was harmed in this experiment.” That language is gone now. The current disclosure says the content “included the use of AI based on templates created, reviewed and edited by journalists in the newsroom.”

The tonal shift suggests the paper recognizes that jokey transparency is not the same as actual accountability. But the change in language also raises a practical question: if journalists are reviewing and editing this content, why are false earthquake reports still live on the site days after the USGS corrected the record?

Automated coverage of earthquakes and wildfires is not an inherently bad idea. Seismic and fire incidents generate enormous amounts of structured data, and there is genuine public value in getting accurate alerts published quickly. The problem at the Bee is not automation itself. It is automation without sufficient human verification at the output stage, particularly when the upstream data sources, including federal agencies with their own automated systems, are themselves prone to error.

Garbage in, garbage out is a principle as old as computing. When the data pipeline runs from an automated federal notification system through a newsroom bot to a published article with no human checkpoint, errors do not just slip through. They get amplified and archived.

The Sacramento Bee covers a region where wildfires and earthquakes are not hypothetical threats. The stakes for getting this coverage right are unusually high. The experiment may continue, but it needs considerably more human oversight before the next fire season arrives.