{"results":[{"item":{"id":"17d799910f0a814d41752df5e559799f6a6eef21","incidentId":"80e9550e1cfdb7cf8c58c26b1d6c465e2e5265c7","title":"When JPEGs Start Giving Orders: A Journey into Multi-modal Prompt Injection | by Jobson | Jul, 2026 | Cyber Security Write-ups","summary":"A security researcher (Jobson) documents discovering multi-modal prompt injection in an AI-powered application that uses a vision-language model to generate captions from user-supplied images or image URLs. After initial SSRF testing failed, the researcher pursued injecting instructions via image content, whose AI-generated captions feed downstream application workflows.","whyItMatters":"Multi-modal prompt injection via user-supplied images shows that vision-language pipelines can be hijacked through image content, threatening any downstream workflow that trusts AI-generated captions.","threatTypeTags":["prompt-injection","multi-modal-prompt-injection"],"affectedTechTags":["llm","vision-language-model","ai-agents"],"threatActor":null,"relevanceScore":0.85,"severityScore":0.5,"sources":[{"sourceId":"firecrawl-search","title":"When JPEGs Start Giving Orders: A Journey into Multi-modal Prompt Injection | by Jobson | Jul, 2026 | Cyber Security Write-ups","link":"https://cybersecuritywriteups.com/when-jpegs-start-giving-orders-a-journey-into-multi-modal-prompt-injection-0b5108f4a7c0"}],"sourceItemIds":["cf4cb758552936ffa97e902eb9c68883f318e193"],"publishedAt":"2026-08-09T07:15:00.012Z","firstReportedAt":"2026-08-09T07:15:00.012Z","curatedAt":"2026-08-09T07:35:18.800Z","itemType":"research","threatStatus":"poc","contentClass":"research","toolPosture":null,"toolCategory":null,"references":[{"role":"aggregator","url":"https://cybersecuritywriteups.com/when-jpegs-start-giving-orders-a-journey-into-multi-modal-prompt-injection-0b5108f4a7c0","domain":"cybersecuritywriteups.com","slug":null,"tier":"unknown","title":"When JPEGs Start Giving Orders: A Journey into Multi-modal Prompt Injection | by Jobson | Jul, 2026 | Cyber Security Write-ups","pageTitle":"When JPEGs Start Giving Orders: A Journey into Multi-modal Prompt Injection | by Jobson | Jul, 2026 | Cyber Security Write-ups"}]},"score":0.26873989068152526},{"item":{"id":"dc66c3fde5b565c7cb6ec9b6cdac26af8be19c0e","incidentId":"ba8292b2ffbf089d6676f0215ea9fade0f902f4f","title":"5 Prompt Injection Examples & How to Defend Each | QAwerk","summary":"QAwerk describes five prompt injection findings its QA engineers discovered while testing live AI-powered products, including a meeting assistant that leaked its own developer instructions, mapping of hidden backends, sentence-completion tricks that bypass refusals, and filter-smuggling techniques, each paired with a defensive mitigation. The piece also frames prompt injection as OWASP's top LLM application risk.","whyItMatters":"Prompt injection in deployed AI products can expose hidden system prompts, backend details, and let attackers override intended behavior, making these tested examples and defenses directly useful for defenders hardening LLM-powered apps.","threatTypeTags":["prompt-injection","jailbreak","system-prompt-leak"],"affectedTechTags":["llm","ai-agents"],"threatActor":null,"relevanceScore":0.75,"severityScore":0.4,"sources":[{"sourceId":"firecrawl-search","title":"5 Prompt Injection Examples & How to Defend Each | QAwerk","link":"https://qawerk.com/blog/prompt-injection-examples/"}],"sourceItemIds":["e5c59dc71c9848d533ce934e3c88eadffaea8857"],"publishedAt":"2026-08-09T07:15:00.012Z","firstReportedAt":"2026-08-09T07:15:00.012Z","curatedAt":"2026-08-09T07:37:04.242Z","itemType":"analysis","threatStatus":"unknown","contentClass":"research","toolPosture":null,"toolCategory":null,"references":[{"role":"aggregator","url":"https://qawerk.com/blog/prompt-injection-examples/","domain":"qawerk.com","slug":null,"tier":"unknown","title":"5 Prompt Injection Examples & How to Defend Each | QAwerk","pageTitle":"5 Prompt Injection Examples & How to Defend Each | QAwerk"},{"role":"link","url":"https://genai.owasp.org/llmrisk/llm01-prompt-injection/","domain":"genai.owasp.org","slug":"owasp-genai","tier":"known","title":"OWASP LLM01: Prompt Injection"}]},"score":0.26707375641998143},{"item":{"id":"86612afd6762c17d4872683b05eec7c255ab76db","incidentId":"bdbb2217dec2ecb80e3a2829647822c112c88ede","title":"What is Prompt Injection? How it Works and How to Prevent It | CloudSEK","summary":"CloudSEK's knowledge-base page titled 'What is Prompt Injection? How it Works and How to Prevent It' is framed as an explainer on prompt injection attacks against LLMs, covering how the attack works and prevention measures. The retrieved page body contained only cookie-consent boilerplate, with no substantive technical content available for analysis.","whyItMatters":"Prompt injection remains a core threat class for LLM-based systems, and educational explainers help defenders understand and mitigate it, though this particular page yielded no analyzable technical detail.","threatTypeTags":["prompt-injection"],"affectedTechTags":["llm"],"threatActor":null,"relevanceScore":0.5,"severityScore":0.1,"sources":[{"sourceId":"firecrawl-search","title":"What is Prompt Injection? How it Works and How to Prevent It | CloudSEK","link":"https://www.cloudsek.com/knowledge-base/prompt-injection"}],"sourceItemIds":["8997b5348553610623134c118a15f21faf1d0c05"],"publishedAt":"2026-07-17T04:45:00.006Z","firstReportedAt":"2026-07-17T04:45:00.006Z","curatedAt":"2026-07-17T05:08:34.879Z","itemType":"analysis","threatStatus":"unknown","contentClass":"analysis","toolPosture":null,"toolCategory":null,"references":[{"role":"aggregator","url":"https://www.cloudsek.com/knowledge-base/prompt-injection","domain":"cloudsek.com","slug":null,"tier":"unknown","title":"What is Prompt Injection? How it Works and How to Prevent It | CloudSEK","pageTitle":"What is Prompt Injection? How it Works and How to Prevent It | CloudSEK"}]},"score":0.2664698799568529},{"item":{"id":"028244b4d01b9ef26ed0984b04c769290f1b3ee1","incidentId":"9249969308f10e91922369e0c3b0e4f6bfc17984","title":"People are using prompt injection to trick Meta's AI into handing over Instagram accounts - Neowin","summary":"Attackers used prompt injection against Meta's AI support assistant on Instagram, sending crafted messages instructing it to link an attacker-controlled email to a target account, causing the AI to send password reset links to the attacker and bypassing 2FA. The exploit was reportedly active in the wild for months, compromising thousands of accounts including a dormant Obama White House account before being patched.","whyItMatters":"It shows that AI agents with account-modifying privileges can be socially engineered via prompt injection to bypass authentication controls and hand over accounts at scale.","threatTypeTags":["prompt-injection","account-takeover","data-exfiltration"],"affectedTechTags":["llm","ai-agents"],"threatActor":null,"relevanceScore":0.95,"severityScore":0.85,"sources":[{"sourceId":"hn-search","title":"People are using prompt injection to trick Meta's AI into handing over Instagram accounts - Neowin","link":"https://www.neowin.net/news/people-are-using-prompt-injection-to-trick-metas-ai-into-handing-over-instagram-accounts/"}],"sourceItemIds":["50f2adde4b016b531a38b4f8765226dfb190a107"],"publishedAt":"2026-06-01T04:13:47.000Z","firstReportedAt":"2026-06-01T04:13:47.000Z","curatedAt":"2026-06-27T04:03:38.550Z","itemType":"incident","threatStatus":"unknown","contentClass":"threat","toolPosture":null,"toolCategory":null,"references":[]},"score":0.2642063166603288},{"item":{"id":"89d0852d491a84e50946aa4bf06c7fc054266754","incidentId":"846ccdf4101fd9b731f79df7a1ebaa31284554d1","title":"A Catalog of Prompt Injection Techniques | Blog | Archestra","summary":"A vendor blog catalogs ten basic prompt injection techniques including context ignoring, fake completion, payload splitting, token smuggling via Base64, few-shot poisoning, defined dictionary attacks, virtualization (grandma trick), DAN jailbreak personas, indirect injection through fetched content, and markdown-image data exfiltration. Each uses a harmless 'I am a sandwich' test string to demonstrate success.","whyItMatters":"Defenders benefit from a clear taxonomy of common prompt injection and jailbreak patterns, especially the agent-targeting indirect injection and markdown-image exfiltration vectors.","threatTypeTags":["prompt-injection","jailbreak","data-exfiltration","indirect-prompt-injection"],"affectedTechTags":["llm","ai-agents"],"threatActor":null,"relevanceScore":0.95,"severityScore":0.4,"sources":[{"sourceId":"hn-search","title":"A Catalog of Prompt Injection Techniques | Blog | Archestra","link":"https://archestra.ai/blog/10-basic-prompt-injections"}],"sourceItemIds":["28e15dc9c685df6a7f84d1e56d43a186bbcb2ecb"],"publishedAt":"2026-06-15T14:00:58.000Z","firstReportedAt":"2026-06-15T14:00:58.000Z","curatedAt":"2026-06-27T04:32:09.204Z","itemType":"research","threatStatus":"unknown","contentClass":"analysis","toolPosture":null,"toolCategory":null,"references":[]},"score":0.26021993038416924}]}