How Prism supported the Karnataka Information Disorder Tackling Unit with continuous multilingual monitoring, signal detection, alerts and information analysis during a three-month pilot.
The Karnataka Information Disorder Tackling Unit operated in an information environment where news, social content, claims, manipulated media and rapidly changing public conversations appeared continuously across multiple languages.
The three-month pilot explored how an intelligence platform could help move from fragmented monitoring toward a more structured understanding of emerging issues.
Prism brought incoming information into a common analytical environment, helping teams identify signals, examine relationships, track sentiment and surface issues requiring further review.
Pilot conducted from 16 March to 16 June 2024.
Alerts surfaced for review across the pilot period.
Multilingual monitoring across major information streams.
Instances where the legal team was notified following identified issues.
The pilot also involved continuous database updates and large-scale sentiment processing. From 3 April, database updates were performed every three hours, with approximately 7,800 articles per update cycle used for sentiment analysis. During 16 May–15 June, the reported average was up to approximately 8,000 articles per day for sentiment analysis.
Signals appeared across different subjects, platforms, languages and formats. A claim could require verification. A video could require deeper examination. A change in sentiment could warrant monitoring. The operational challenge was connecting these individual observations into a usable picture.
The alerts generated during the pilot illustrate that information disorder is not a single category of problem. Alerts ranged from potentially inflammatory content and misleading claims to edited videos, alleged incidents and content requiring immediate verification.
Several alerts involved claims or media where the immediate value of detection was to trigger fact-checking, source verification or continued monitoring rather than to declare the content false automatically.
Individual posts could appear isolated. When viewed within the broader information environment, however, issues could become more relevant to public communication, law and order, political discourse or institutional reputation.
The pilot included video and image-related cases where identifying an item was only the first step. Further review was required to establish whether content had been edited, misrepresented or otherwise required intervention.
Prism surfaced and structured signals. Subsequent fact-checking, legal review, clarification or other action remained a human and institutional decision.
The pilot reporting records 319 total alerts. A separate severity breakdown records 183 High, 115 Moderate and 17 Low alerts, which totals 315. The discrepancy is preserved here rather than presenting the severity distribution as a reconciled 319-alert breakdown.
Across the pilot, the alerts demonstrated several recurring information patterns that matter to institutions operating in high-stakes environments.
Claims involving public figures, incidents, institutions and political developments frequently required rapid verification before their significance could be established.
Some cases involved videos whose apparent meaning could change materially when the underlying footage or editing was examined.
Certain alerts involved content with potential communal, caste or public-order implications, making timely review particularly important.
Alerts also surfaced claims involving elected representatives, government institutions and public officials that warranted verification or continued monitoring.
The following examples illustrate the range of signals surfaced during the pilot. They are presented as examples of detected issues requiring review—not as evidence that Prism itself determined the ultimate truth or caused subsequent action.
An item was surfaced for potential communal hate implications. Immediate fact-checking was recommended.
Two videos involving alleged casteist slogans were surfaced. Continued monitoring and verification were recommended.
A circulating claim alleging that a woman was chased by individuals in Karnataka was surfaced for immediate verification.
A claim involving Congress MLA Raju Kage was surfaced. Review found the circulated video to be edited, highlighting the importance of examining source media before drawing conclusions.
A claim concerning an alleged security breach involving the Deputy Chief Minister was surfaced for immediate verification.
A viral Instagram post alleging an attack on an ambulance was surfaced and marked for immediate fact-checking.
The pilot monitored information across Kannada, English, Hindi, Marathi, Tamil, Telugu, Tulu and Urdu.
The objective was not simply to translate information. It was to bring signals from different linguistic environments into a common analytical workflow so that relevant developments could be reviewed together.
Information could be monitored continuously rather than examined only after a development became prominent.
Potentially important changes, claims and information patterns could be surfaced for further examination.
Sentiment analysis and temporal views helped teams understand how information environments changed over time.
Detected issues could be brought into a workflow for verification, monitoring and institutional review.
Video and image-related information could be surfaced for deeper examination, including cases involving edited media.
Search and relationship-oriented analysis helped move beyond individual items toward a broader information picture.
Large volumes of incoming information create visibility, but the operational value comes from identifying the signals that deserve attention.
Alerts created a starting point for fact-checking, investigation and institutional review rather than replacing those processes.
Monitoring across eight languages allowed information from different linguistic environments to be considered within the same operational picture.
Claims, videos and posts cannot always be understood from the individual item alone. Their surrounding information environment can materially change their significance.
Prism helped surface and structure information. Decisions about verification, communication, legal review or other responses remained with the responsible teams.
During the Karnataka deployment, Prism processed up to approximately 64,000 articles per day across the monitored information environment.
Within this larger processing pipeline, approximately 7,800 English-language articles per day were processed for sentiment analysis.
The approximately 7,800 articles/day figure refers specifically to the English-language sentiment-analysis workload and should not be interpreted as Prism's total daily article-processing volume.
The Karnataka pilot demonstrated the value of moving beyond simply collecting information. In a multilingual, continuously changing environment, the ability to surface signals, structure alerts, examine context and connect detection with human review can help institutions focus attention where it matters most.
See how Prism can help identify the signals, narratives and emerging pressure within yours.
Request a Demo →