Perplexity Context API adds Decider safety filtering
Following feedback from Shopify CEO Tobi Lütke, Perplexity integrated its Decider models into the newly launched Context API to enable safe web fetching for autonomous agents. Developers can now evaluate fetched pages against custom rubrics to detect prompt injections and filter untrusted content before ingestion.
Unfiltered web retrieval is the single largest attack vector for autonomous agents, making probabilistic safety filtering table stakes for production agent workflows. Perplexity shipping integrated prompt injection checks within twenty-four hours highlights how quickly agent infrastructure is hardening against indirect injection attacks.
- –Indirect prompt injection remains an unsolved vulnerability at the model layer, requiring strict retrieval-time gatekeepers before web content enters agent memory.
- –Decider returns structured numerical probabilities rather than conversational text, letting developers enforce strict rejection thresholds without brittle regex parsing.
- –Embedding safety scoring directly into the Context API fetch pipeline removes the latency and cost overhead of routing content through separate moderation services.
- –Rapid turnarounds spurred by high-profile builders like Tobi Lütke demonstrate Perplexity's focus on securing the default data retrieval layer for autonomous agents.
DISCOVERED
37m ago
2026-11-19
PUBLISHED
2h ago
2026-11-19
RELEVANCE
AUTHOR
AravSrinivas