Can EXIF data be trusted? Dates, camera models and GPS can be edited

Not on its own. EXIF is a set of fields that software writes into an image file, and other software can change or remove them. A date, a camera name or a location in EXIF is a label, not proof. A signed record differs: it shows whether the file still matches what was signed, not whether its claims are true.

What is EXIF data?

EXIF is a standard for technical details stored inside a photo file. The Camera & Imaging Products Association (CIPA) publishes it as DC-008, Exif Version 3.1. Common fields include when a picture was taken, what camera made it, and where. Each is a field in the file, like a caption on the back of a print.

Software writes each field: a camera, a phone, an editing program or a website. Software can later rewrite it.

Can EXIF data be edited?

Yes. The IPTC Photo Metadata User Guide describes metadata as data embedded in the image file that photo software can read and edit. We read it on October 30, 2026. Three points matter here.

• Fields are shared and rewritten. The guide says the IPTC Date Created field can be derived from the Exif DateTimeOriginal field if the software supports it, and that changing the IPTC field can also change the Exif tag, depending on the software. So a date can move without anyone meaning it to.
• Software fills gaps. The guide says Exif does not hold time zone information in its time stamp, and that most software applies the local time zone of the receiving computer when it imports the time. It tells readers to check this when an image was made in a different time zone.
• Systems strip fields. The guide warns that a site's content or image management system or content delivery network may be stripping out embedded fields. It says some systems do this to save a few bytes of bandwidth.

None of this needs bad intent. Ordinary software changes and drops metadata, and a person can do the same on purpose. We do not give steps for that.

Why is EXIF a label and not proof?

In an ordinary file, an EXIF field is not covered by a signature, so nothing checks it against the pixels. If the image changes, the field does not notice. If the field changes, the image does not notice. The C2PA explainer makes the contrast from the other side. Section 7.4.2 says a person can place standard metadata inside a Content Credential, and that this makes it tamper-evident. We read that as meaning metadata left outside a credential is not covered. Section 7.1.3 says that any change to the asset or to the provenance, however small, changes the hashes, so the two stop matching. Section 7.2.3 answers the question "Can the provenance metadata be removed?" with "Yes it can."

A signed record is not proof of truth either. The explainer says Content Credentials make no value judgment on whether the provenance data is true (section 2), and that provenance information alone cannot tell you whether the content is true, accurate or factual (section 7.2.2). A signed record tells you who signed, if a trust list vouches for the signer, and whether the bytes still match. It does not make the claim inside it correct.

What did our checker do with metadata labels?

On October 30, 2026 we ran the checker in Chrome on 32 real files and on files we made. The 32 real files are 29 from Wikimedia Commons categories named "Images generated by" a product, 2 vendor samples and 1 C2PA test image. Third parties uploaded the Commons files, and we cannot know how each was handled before upload. They show what the checker reads in files we hold, not what any product does.

The test picture says, in unsigned metadata, that it is a digital capture. The checker reports "Metadata declares another source" and the line "A label, not proof." The real Ideogram file names a tool in its EXIF Software field. The checker reads only "Tool name found in metadata", a hint, not a declaration. In both cases it cannot tell a truthful label from an untruthful one. Only a signed record, checked against a trust list, would give it more to go on.

Across the 32 real files, 5 read "Metadata found; source not established" and 1 reads "Tool name found in metadata". Two of those five are the JPEGs in the table, with an EXIF date or EXIF dates. A date in a file filed under a product's name shows only that a date field exists.

How does a signed record differ in a test?

We also changed 12 bytes of image data in a signed Firefly sample and in a signed Microsoft sample. Both read "Content Credentials failed validation", because the pixels no longer matched the record. An EXIF label has no such check.

A signed record is only as good as its signer. We signed a picture we drew with a self-made certificate. The checker reads it as "Content Credentials found; signer trust not established", and says that anyone can make a certificate with any name. Can Content Credentials be faked? covers that case in full.

What this means when you check an image

Open the checker, choose the file, and look at the headline. These two headlines come from unsigned labels.

• "Metadata declares another source" means unsigned metadata says the image came from a non-generative source, such as a camera capture. It is a label, not proof, and anyone with ordinary software can add or remove it.
• "Tool name found in metadata" means a field, such as the EXIF Software field, names a tool. It is a hint. It does not say how the image was made.

Neither headline tells you how the image was made. The limits line always applies: "These checks cannot determine whether an image is AI-generated." If a date, a camera or a place matters to a decision, ask the sender for the original file.

Guides on this topic

• How to tell if an image is AI is the main page on this topic.
• Content Credentials versus EXIF, IPTC and XMP compares a signed record with unsigned labels.
• Where AI declarations live in an image file shows the places a label can sit.
• How to view image metadata covers reading the fields.
• Can Content Credentials be faked? covers signed records and self-made certificates.

Frequently asked questions

Can GPS data in a photo be wrong?

Yes. It is a field in the file, like the date and the camera name, and software can write or change it. Treat a location in metadata as a claim.

Does a missing EXIF date mean a picture was edited?

No. The IPTC guide warns that some systems strip embedded fields, and says some do it to save bandwidth. A missing field has many possible causes and none is visible from the file alone.

Is EXIF useless, then?

No. It is a useful clue and a place to start. It is not evidence by itself.

Can a signed record be edited as well?

A change to the image after signing, or damage to the record, fails validation: in our tests both read "Content Credentials failed validation". The record can also be removed, so its absence says little.