Meta builds an AI‑detector for Meta AI
Meta is developing an AI Detector for its Meta AI product, aiming for automated synthetic-content identification amid rising deepfake and synthetic-content concerns. The report frames the tool as part of a broader response to authenticity challenges in large-scale AI deployments reported.
TestingCatalog’s discovery of an in‑product “AI Detector” for Meta AI was first flagged in a TestingCatalog post and covered by outlets including Gadgets360. (testingcatalog.com) Public coverage notes the detector’s modality scope remains unspecified—reports say it’s unclear whether the tool targets text, images, or video. (gadgets360.com) Meta’s existing approach already embeds invisible watermarks and IPTC metadata into images produced by Meta AI, and the company said it uses those signals to apply “Imagined with AI” labels. (about.fb.com) Meta has stated it’s collaborating with industry partners and forums such as the Partnership on AI to agree on common technical standards for identifying synthetic media. (partnershiponai.org) Company statements and reporting indicate engineers are building detection systems intended to work even when invisible markers are absent, and Meta has trialed LLMs for automated moderation tasks. (sammobile.com) Security researchers and journalists warned that AI‑detection systems are often evadable in practice, with at least one expert quoted calling such detectors “easily evadable” and noting limits for audio and video. (independent.co.uk) When Meta announced broader labeling in February 2024, Nick Clegg said labels would be applied “in the coming months,” and the company outlined a timeline that included shifting policy and rollout steps through mid‑2024. (about.fb.com)
