Searchlight Cyber researchers have disclosed a critical zero-day vulnerability in WordPress, a pre-authentication remote code execution flaw reportedly valued at up to five hundred thousand dollars by exploit brokers. The discovery, completed in under ten hours at a compute cost of approximately twenty-five dollars, was executed by a specialized AI model designated as GPT5.6 Sol Ultra. The finding impacts millions of globally deployed WordPress instances and signals a rapid evolution in automated security research methodologies. The vulnerability originates in WordPress’s batch API, a feature engineered to process multiple virtual API requests within a single payload. The AI agent identified a synchronization fault in how validation checks and execution handlers are sequenced. By intentionally desynchronizing these arrays, the model bypassed standard parameter sanitization and permission validation. This architectural mismatch was then chained to a GET endpoint that filters posts by author exclusion. The desynchronization allowed malicious SQL syntax to bypass sanitization routines, resulting in a fully functional pre-authentication SQL injection. Transforming the injection into full remote code execution required a complex multi-stage chain. The AI leveraged WordPress’s in-memory caching system to poison post data, triggering a reconciliation process between cached objects and database records. By manipulating oEmbed cache entries and forcing the system to update a fabricated post, the agent exploited a cycle-detection mechanism within the post hierarchy. This specific logic path inadvertently triggered a filter that reconciled conflicting cached representations without overriding content fields, enabling the injection to alter the post type. The modified entry masqueraded as a customize_changeset, a WordPress structure used to save theme and configuration drafts. When the system processed this crafted entry, it temporarily elevated the request context to administrator privileges based on the changeset’s embedded user metadata. Leveraging this temporary elevation, the exploit hijacked WordPress’s dynamic hook system. By constructing a post object with a fabricated status and type combination, the agent triggered the parse_request action hook. This action replayed the original malicious batch request under the assumed administrator identity, successfully provisioning a new admin account and deploying a backdoor for full system access. The researcher highlighted that the exploit’s interconnectivity, spanning API parsing, database caching, theme configuration handlers, and dynamic hook dispatching, would be exceptionally difficult to construct manually within a comparable timeframe. The discovery reinforces the growing capability of large language models to map disparate codebases, identify cross-module weaknesses, and synthesize complete exploitation paths. Searchlight Cyber responsibly disclosed the flaw to the WordPress security team prior to publication, allowing developers to deploy patches before public technical details circulated. The company continues to monitor the threat landscape and anticipates an urgent security advisory to mitigate widespread exploitation across WordPress’s extensive global infrastructure.