X has deployed a new version of its Grok AI model that detects duplicated content at three times the rate of its predecessor, using it as the technical backbone for a broader crackdown on creators who republish stolen posts for ad revenue. The platform detected 1.5 million stolen posts in its latest enforcement cycle as of July 2026, according to TechCrunch.

Under the new policy, X will redirect monetized impressions to the original uploader when stolen content has been disguised with watermarks, intros, or other edits. The detection and payout-redirection mechanism covers copies of viral text posts as well as video. Over $1 million in creator payouts will be redirected to the original creators of stolen content under the new framework.

The enforcement system operates on multiple tiers. Users caught in repeated or intentional attempts to get around the anti-content-theft policy will be removed from the creator revenue-sharing program entirely. X also maintains a three-strike policy for engagement bait — content designed to trick users into interacting through deceptive tactics. Users caught three or more times are removed from the creator program and forwarded to the policy team for account suspension. These measures run alongside X's ongoing bot-removal infrastructure, which was identifying and suspending 208 bots per minute as of April 2026.

X product executive Nikitabier publicly criticized top creator Mr. Beast for using financial bait to drive video engagement, signaling that the platform's enforcement posture extends beyond anonymous bad actors to high-profile creators whose tactics generate significant platform traffic.

The technical challenge here is real. Catching near-duplicate video that has been lightly modified with intros, watermarks, or re-encoding requires perceptual hashing or embedding-based similarity matching at scale, not exact-match fingerprinting. Perceptual hashing works by analyzing the visual or audio characteristics of a file — the way it looks or sounds — rather than the underlying data itself, so that a re-encoded or lightly edited copy still produces a similar hash. Applying the same detection to viral text posts adds a different problem: paraphrased or reposted text that may not match word-for-word but is substantively the same content. Grok's threefold improvement in detection rate suggests X has moved from simpler hash-based matching toward semantic similarity models, which compare the meaning of content rather than its exact wording, and can therefore catch derivative variations.

The payout-redirection mechanism is the more novel component. Content-ID systems on platforms like YouTube have long identified duplicates, but redirecting monetized impressions to the original uploader rather than simply demonetizing or removing the copy changes the economic calculus for content theft. A bad actor who previously profited from repackaging viral posts now generates revenue for the original creator instead. Whether the detection accuracy is sufficient to avoid false positives at scale is an open question, and one that will matter considerably for creators whose original content is misidentified as derivative.

The broader context here is a platform attempting to align its creator economy with content provenance — verifying who originally created a piece of content. X's revenue-sharing program, launched after the platform's rebrand from Twitter, created financial incentives for engagement that also incentivized content theft, engagement bait, and bot-driven amplification. The current set of measures, taken together, is an attempt to close that loop: better detection, economic redirection, escalating penalties, and bot suspension operating in concert. The $1 million-plus in redirected payouts is a modest sum relative to the total creator economy on the platform, but it establishes a mechanism whose scope could expand as detection improves.

For platform engineers and trust-and-safety teams, the architecture is familiar in shape if not in specifics: a detection model feeding an enforcement pipeline that escalates from economic penalty to program removal to account suspension. The interesting variable is the false-positive rate at the payout-redirection stage, where a misattribution directly transfers money between creators. That is a higher-stakes error than a post being labeled as duplicate and removed, and it is one X will need to handle transparently if the system is to retain creator trust.

The Nikitabier-Mr. Beast dynamic is worth noting. Public criticism of a top creator by a platform executive signals that engagement-bait enforcement will not be limited to accounts with limited reach. If the policy applies evenly, it could reshape incentive structures for creators who have built large followings partly through financial-bait tactics. If enforcement skews toward smaller accounts while high-traffic creators receive warnings rather than strikes, the policy's credibility erodes quickly.