MobileInject: Evaluating Reasoning-to-Action Security of MLLM-Based Mobile Agents under Cross-Modal Consistency Injection
Evaluates reasoning-to-action security of MLLM mobile agents under cross-modal injection.
AI Storyline
3 items · 2 sources · 3 days
Operational story trace
Security Injection
Follow in this browser to see new updates on your Live feed.
Current stateDeveloping
Latest change
SecPO (Oct 30) proposes adversarial training against prompt injection, positioned against Meta-SecAlign, the existing best-defended LLM.
Earlier contextThe story so far
MobileInject (Oct 16) framed prompt injection as a reasoning-to-action risk for MLLM-based mobile agents. Heads of Security (Oct 27) then proposed a defense: learned per-head attention gates that enforce instruction-data separation.
editor-curated · source-linked
Arc
Oct 16Oct 30 · now
THE ATTACK · Oct 16MobileInject evaluates injection against mobile agents1 source · show source ▾
DEFENSE · Oct 27Heads of Security gates attention to separate instructions from data1 source · show source ▾
Heads of Security: Learned Attention Gates Improve Prompt Injection Robustness
Introduces learned per-head attention gates as an instruction-data separation defense.
NOW · Oct 30SecPO applies principled adversarial training1 source · show source ▾
SecPO: Principled Adversarial Training for Prompt Injection Security
Adds adversarial training aimed at beating Meta-SecAlign, the current best-defended LLM.
Day 1 Friday, Oct 16, 2026
MobileInject: Evaluating Reasoning-to-Action Security of MLLM-Based Mobile Agents under Cross-Modal Consistency Injection
Evaluates reasoning-to-action security of MLLM mobile agents under cross-modal injection.
Day 2 Tuesday, Oct 27, 2026
Heads of Security: Learned Attention Gates Improve Prompt Injection Robustness
Introduces learned per-head attention gates as an instruction-data separation defense.
Day 3 Friday, Oct 30, 2026
SecPO: Principled Adversarial Training for Prompt Injection Security
Adds adversarial training aimed at beating Meta-SecAlign, the current best-defended LLM.
What to watch — open questions
- Do HoS or SecPO results hold against adaptive attackers outside their own evaluations?
- Will either defense ship in an open-weight model or an agent framework?
How this thread was built
editor wrote the arc · 3 beatswatcher 1 status change
Storylines are threaded mechanically from the feed: stories that share a distinctive anchor across multiple days and sources. Each item links to its original source. The evidence trace, current state, and open questions are written by the editor routine and refreshed whenever a new beat lands.