Answer: Start with trusted sites and avoid unfamiliar websites or those with user-generated content from unknown sources. Switch to "Manually approve" to review every action, confirm before sensitive or high-risk tasks, and stop immediately if Claude discusses unrelated topics, accesses unexpected websites, or requests sensitive information, as this could indicate a prompt injection attempt. Status: Confirmed Entry: tool/claude-apps/permissions-and-security/protect-against-prompt-injection Revision: 3 Digest: sha256:5c26e0c4677f4d25 (changes whenever the content changes) Timeline: Created: 2026-11-13T02:58:19Z Last changed: 2026-11-19T11:59:31Z (suspicious behavior indicator: — → prompt injection attempt) Last verified: 2026-11-20T12:02:56Z Next check: 2026-11-21T12:02:56Z Check interval: Daily Expected change: No fixed schedule Data: Field | Value | Source | As of trusted sites guidance | Begin with websites you trust | [1] | 2026-11-20 default permission mode | Automatically approve | [1] | 2026-11-20 recommended review mode | Manually approve | [1] | 2026-11-20 suspicious behavior indicator | prompt injection attempt | [1] | 2026-11-20 reporting method | in-chat feedback options | [1] | 2026-11-20 Sources: [1] Anthropic. "Use Claude in Chrome safely | Claude Help Center". https://support.claude.com/en/articles/12902428-use-claude-in-chrome-safely Role: primary. Accessed: 2026-11-20T12:02:56Z Quote: "Start with trusted sites: Begin with websites you trust." (Supports: trusted sites guidance) Quote: "The Cowork side panel defaults to "Automatically approve" mode" (Supports: default permission mode) Quote: "Switch to "Manually approve" if you want to review every action" (Supports: recommended review mode) Quote: "This could indicate a prompt injection attempt." (Supports: suspicious behavior indicator) Quote: "flagging any concerning behavior through the in-chat feedback options." (Supports: reporting method) Snapshot SHA-256: af2bea06dead88edf1f542e2f20fe9037c19063b8e6ce92d993f0fc420b5794e (captured 2026-11-19T11:59:31Z) Verification: Method: Automated extraction from [1] with deepseek-flash (JSON output, thinking disabled). Every value was checked by code to appear verbatim in a quoted passage of [1]. On this check the text of [1] was unchanged since 2026-11-19T11:59:31Z, so the values verified then were kept. Cross-check: not-available at 2026-11-20T12:02:56Z. No independent second source is configured for this entry Last successful check: 2026-11-20T12:02:56Z Last attempt: 2026-11-20T12:02:56Z Notes and limitations: Extracted automatically from a single official source (section "Protecting yourself from malicious attackers"). Related: https://latestn.com/tool/claude-apps/permissions-and-security/remove-group-access-shared-resources How do I remove a group's access to shared projects, skills, and plugins in Claude? https://latestn.com/tool/claude-apps/permissions-and-security/report-safety-issue How do I report a safety issue to Claude? https://latestn.com/tool/claude-apps/permissions-and-security/report-security-vulnerability How do I report a security vulnerability in Claude or Anthropic systems? https://latestn.com/tool/claude-apps/permissions-and-security/permissions-combine-multiple-roles How do permissions combine when a Claude Enterprise member has multiple custom roles? https://latestn.com/tool/claude-apps/permissions-and-security/pre-approve-commands How do I pre-approve common commands in Claude Code permissions? https://latestn.com/tool/claude-apps/permissions-and-security/prohibited-agent-uses What agent uses are prohibited under Claude's Usage Policy? https://latestn.com/tool/ All entries in AI tools Change log: Time | Event | Change | Source 2026-11-19T11:59:31Z | changed | suspicious behavior indicator: — → prompt injection attempt | [1] 2026-11-19T11:59:31Z | changed | recommended review mode: — → Manually approve | [1] 2026-11-19T11:59:31Z | changed | default permission mode: — → Automatically approve | [1] 2026-11-19T11:59:31Z | changed | reporting method: flagging any concerning behavior through the in-chat feedback options → in-chat feedback options | [1] 2026-11-19T11:59:31Z | changed | prompt injection indicator: This could indicate a prompt injection attempt → — | [1] 2026-11-19T11:59:31Z | changed | suspicious behavior signs: If Claude suddenly starts discussing unrelated topics, accessing unexpected websites, or requesting sensitive information, stop the task immediately → — | [1] 2026-11-19T11:59:31Z | changed | manual approval mode: Manually approve → — | [1] 2026-11-19T11:59:31Z | changed | sites to avoid: Avoid unfamiliar websites or those containing user-generated content from unknown sources → — | [1] 2026-11-17T12:04:15Z | changed | suspicious behavior signs: — → If Claude suddenly starts discussing unrelated topics, accessing unexpected websites, or requesting sensitive information, stop the task immediately | [1] 2026-11-17T12:04:15Z | changed | manual approval mode: — → Manually approve | [1] 2026-11-17T12:04:15Z | changed | sites to avoid: — → Avoid unfamiliar websites or those containing user-generated content from unknown sources | [1] 2026-11-17T12:04:15Z | changed | trusted sites guidance: — → Begin with websites you trust | [1] 2026-11-17T12:04:15Z | changed | reporting method: in-chat feedback options → flagging any concerning behavior through the in-chat feedback options | [1] 2026-11-17T12:04:15Z | changed | starting guidance: Begin with websites you trust → — | [1] 2026-11-17T12:04:15Z | changed | default permission mode: Automatically approve → — | [1] 2026-11-17T12:04:15Z | changed | permission mode for review: Manually approve → — | [1] 2026-11-17T12:04:15Z | changed | suspicious behavior action: stop the task immediately → — | [1] 2026-11-13T02:58:19Z | added | Entry created | — Changes for this entry as a feed: https://latestn.com/tool/claude-apps/permissions-and-security/protect-against-prompt-injection/latest Challenge this entry: Open challenges: None How: If a value here is wrong or out of date, send what is wrong and a link to evidence. Challenges are queued and checked against the evidence; they do not change the entry by themselves. If the entry changes as a result, the change log says so. Form: https://latestn.com/dispute/tool/claude-apps/permissions-and-security/protect-against-prompt-injection POST: https://latestn.com/dispute/tool/claude-apps/permissions-and-security/protect-against-prompt-injection with claim, evidence_url, and optionally field and contact (form-encoded or JSON) GET: https://latestn.com/dispute/tool/claude-apps/permissions-and-security/protect-against-prompt-injection?claim={what is wrong}&evidence_url={link} How to cite: LatestN. (2026). How do I protect myself from prompt injection attacks when using Claude in Chrome? Revision 3, last verified 2026-11-20T12:02:56Z. https://latestn.com/tool/claude-apps/permissions-and-security/protect-against-prompt-injection (accessed 2026-11-21T10:05:29Z). @misc{LatestN_tool_claude_apps_permissions_and_security_protect_against_prompt_injection_r3, title = {How do I protect myself from prompt injection attacks when using Claude in Chrome?}, author = {{LatestN}}, year = {2026}, howpublished = {\url{https://latestn.com/tool/claude-apps/permissions-and-security/protect-against-prompt-injection}}, note = {Revision 3; digest sha256:5c26e0c4677f4d25; last verified 2026-11-20T12:02:56Z; accessed 2026-11-21T10:05:29Z} } Reuse: License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) Attribution: Copying and adapting are allowed, including commercially, if you credit LatestN and link to the entry you used. Each entry ends with a ready-made citation. URL: https://latestn.com/tool/claude-apps/permissions-and-security/protect-against-prompt-injection