Knowledge cutoff date is disclosed where relevant

Language models have a training cutoff and cannot know about events, legislation, software releases, or market conditions that postdate it. When an application provides no disclosure — neither in the system prompt nor the UI — users treat AI responses as current, leading to decisions based on outdated information. For compliance-sensitive domains (GDPR enforcement updates, securities regulations, medical guidelines), acting on stale AI output can cause real harm. NIST AI RMF GOVERN-1.1 requires transparency about AI system limitations. OWASP LLM09 identifies temporal misinformation as an LLM risk category.

Add a cutoff disclosure to the system prompt with the model's specific training cutoff date:

const systemPrompt = `
Your knowledge cutoff is [MODEL_CUTOFF_DATE]. For questions about recent events,
current prices, active legislation, or software versions, proactively note that
your information may be outdated and direct the user to verify with a current source.
`

Alternatively, add a static footer or tooltip in the chat UI — for example, src/components/chat/ChatPanel.tsx — displaying: "AI responses reflect training data through [date]. Verify time-sensitive information independently."

• ID: ai-response-quality.source-attribution.knowledge-cutoff-disclosure

• Severity: high

• What to look for: Enumerate all relevant files and Check the system prompt for any mention of the model's training cutoff date, or for instructions directing the AI to proactively disclose when a question involves current events, recent data, or time-sensitive information that may be outside its training window. Check whether the application's UI includes any static disclosure about AI knowledge limitations (e.g., "This AI's knowledge has a cutoff of [date]" in a footer or info tooltip).

• Pass criteria: At least 1 implementation must be present. Either (a) the system prompt instructs the AI to proactively flag time-sensitive topics with a knowledge cutoff caveat, or (b) the application UI displays a static disclosure about the AI's knowledge limitations.

• Fail criteria: No cutoff disclosure mechanism exists in either the system prompt or the UI for an application that answers questions about real-world facts, events, or current information.

• Skip (N/A) when: Application is strictly a code assistant, document summarizer, or similar tool where knowledge currency is irrelevant to the use case.

• Detail on fail: "No knowledge cutoff instruction in system prompt and no static UI disclosure — users may not know AI responses could be outdated" (max 500 chars)

• Remediation: Add a cutoff disclosure to your system prompt and/or UI:

  const systemPrompt = `
  Your knowledge has a cutoff date of [MODEL_CUTOFF_DATE]. For questions about
  recent events, laws, prices, software versions, or anything that may have changed,
  proactively note that your information may be outdated and suggest the user verify
  with a current source.
  `
  

  Alternatively, add a visible note in the chat UI: "AI responses are based on training data with a knowledge cutoff of [date]."