ROOKVEYL INTELLIGENCE EXCHANGE

Agent security & prompt injection — research landscape

Every lot is frozen at listing time with its sources, license, and retrieval record. The winner pays once—never a subscription.

Market onlineLive checkout ready
Pay once if you win

Bid in USD. Only the winner pays through Stripe after the deadline. Confirmed payment unlocks the saved JSON packet and receipt.

REAL-MONEY BIDDING
LIVEOpen

LIVE source data

20 research records with publication and publisher charts. Metadata analysis, not executable algorithms. One stored edition, one winner.

What you receive

20 records in a downloadable JSON packet, with descriptive analysis, source links, licensing notes, and a payment receipt.

Build a source-linked reading list for an agent threat model and security evaluation.

One saved edition · no subscription · public sources remain nonexclusive
Opening bid$25
Bid activity0bids
ClosesSep 18, 3:35 PM UTC15h 44m left
20 source records Frozen snapshot $5 increment
Preview contents and provenance
Coverage in this packet

Descriptive counts of the providers actually present in this packet, publication dates and title terms. Collection freshness does not imply a new publication. Social/news inputs contribute aggregate context only; this is not sentiment truth, investment advice, or a quality ranking.

20 research records0 context signals1 source families
arxiv20
injection 7prompt 7jailbreak 6attacks 5jailbreaks 5against 3evaluation 3indirect 3
2026-087
2026-0913
01
TACS: Trajectory-Aware Candidate Selection for LLM Jailbreak Suffix OptimizationarXiv · retrieved Sep 16, 3:35 PM UTC
02
Reachability-Based Capability Confinement for LLM Agents under Indirect Prompt InjectionarXiv · retrieved Sep 16, 3:35 PM UTC
03
Will the User Ever Know? Covert Indirect Prompt Injection Attacks on Tool-Using LLM AgentsarXiv · retrieved Sep 16, 3:35 PM UTC
04
ECLIPSE: Self-Evolving Stealthy Prompt Injection Attack against Long-Horizon Agentic SystemsarXiv · retrieved Sep 16, 3:35 PM UTC
05
The Fragility of Jailbreak Robustness Across Operational StatesarXiv · retrieved Sep 16, 3:35 PM UTC
06
Validity-Aware Jailbreak Evaluation for Large Language ModelsarXiv · retrieved Sep 16, 3:35 PM UTC
07
HiveTraceGuard-Pro: A Compact Generative Guardrail for Prompt Injection, Jailbreaks, and Adversarial ObfuscationarXiv · retrieved Sep 16, 3:35 PM UTC
08
Jailbreaking Text-to-Image Models Through Cracks: Navigating Heterogeneous Safety Filters via Multi-Agent DebatearXiv · retrieved Sep 16, 3:35 PM UTC
09
Context Inference Attacks Without JailbreaksarXiv · retrieved Sep 16, 3:35 PM UTC
10
Breadth Beats Depth: Improving GCG-Based Jailbreak Optimization with Breadth-Oriented Suffix SearcharXiv · retrieved Sep 16, 3:35 PM UTC
11
Before the Script, Set the Stage: How Worldview Simulation Amplifies Psychologically Grounded Persuasion in Multi-Turn JailbreakingarXiv · retrieved Sep 16, 3:35 PM UTC
12
AlcaTRAz - Anchored Tree-Rule Defense Against JailbreaksarXiv · retrieved Sep 16, 3:35 PM UTC
13
IndicSafeEval: Safety Robustness of Large Language Models under Multilingual Persuasive Jailbreak AttacksarXiv · retrieved Sep 16, 3:35 PM UTC
14
Rethinking Indirect Prompt Injection as a Test-Time Search ProblemarXiv · retrieved Sep 16, 3:35 PM UTC
15
Repeat-After-Me: Black-Box Adaptive Visual Prompt InjectionarXiv · retrieved Sep 16, 3:35 PM UTC
16
Beyond the Verdict: Evidence-Aligned Evaluation of Visual Prompt-Injection GuardrailsarXiv · retrieved Sep 16, 3:35 PM UTC
17
Bait-and-Recover: Poisoning Internal Refusal Signals to Defend LLMs against White-Box Editing JailbreaksarXiv · retrieved Sep 16, 3:35 PM UTC
18
SAFEGuard: Detect Optimization-Based Jailbreak Attacks Through Harmful Semantic Analysis and Fluency MeasurementarXiv · retrieved Sep 16, 3:35 PM UTC
19
Structural Jailbreaks Generalize but Do Not Compound: A cross-provider and multilingual study of Involuntary In-Context LearningarXiv · retrieved Sep 16, 3:35 PM UTC
20
An Experimental Evaluation of Multimodal Prompt Injection Attacks on Agentic AI FrameworksarXiv · retrieved Sep 16, 3:35 PM UTC

One buyer receives this analysis edition. Source metadata remains public and nonexclusive. No paper text, algorithm implementation or patent rights included.

Questions for your evaluation
  • What attacker access and trusted boundaries does each paper assume?
  • What benign-task failures accompany a proposed defense?
  • Which claims can you reproduce safely in an isolated test environment?

A research shortlist and descriptive analysis. Scientific claims and performance have not been independently reproduced. Public source material remains nonexclusive.

Included fields: title, sourceUrl, retrievedAt, published, publisher, abstract, authors, tags, verification, rights.

SHA-256 048219540cb7fa5f611423f88c88b4318f40b60846818b105f728ab0f6841ba3

Fixed rules. Highest accepted bid at the deadline wins. Your authorized agent may bid within its limit. No automatic card charges or deadline extensions. Public-source information remains nonexclusive.