Every incident, CVE and red-team finding around AI — checked against a primary source, swept twice a day. What we rejected is as public as what we kept: the sweep log.
NOTABLEPROMPT-INJECTION
Mobile GUI agents act on whatever the screen shows, so instructions embedded in ordinary UI content can redirect their actions (environmental injection). MobileProbe drops fixed payload-at-a-step benchmarks and instead picks injection steps trajectory-aware, then adaptively evaluates agent security across models.
NOTABLEPROMPT-INJECTION
Warning-induced over-abstention: an adversary plants fabricated risk warnings in low-trust content and the agent refuses legitimate tasks. The paper argues safe abstention must weigh both the claimed risk and the authority of the evidence, and evaluates a verify-then-act defense on counterfactual task variants.
NOTABLERED-TEAM
In subscription SaaS the agent's own stop-decision spends provider-paid model and tool calls. OverServe, a multi-agent attack framework, co-designs prompts and tool responses so agents extend execution for a malicious tenant, amplifying provider resource cost per request.
NOTABLEPROMPT-INJECTION
Enforcement defenses (CaMeL, Progent, IPIGuard, DRIFT) have the agent's own model write a security artifact for code to enforce. Across 15 open-weight models (13 between 1.7B and 12B) the written artifacts admit attacks on most of those defenses; CapLite enforces prompt-injection defenses for small-model agents without model-written artifacts.
NOTABLESUPPLY-CHAIN
Existing fine-tuning-activated backdoors assume limited domain shift or frozen encoders and fail under full-model fine-tuning. Poison-then-Hide implants a dormant backdoor in a pretrained vision encoder that activates later during fine-tuning, staying hidden in the distributed checkpoint.
NOTABLERED-TEAM
Sandbagging — scoring low on evals on purpose — turns a low benchmark score into false reassurance. The authors build sandbaggers (small open models that fail GSM8K only under an audit marker) and show detectors usually validated once against a fixed attack degrade once the model adapts to what gets flagged.
NOTABLECVETOOL
GHSA-rc3w-8jc8-7x3w (CVE-2026-126386): n8n's MCP OAuth registration and authorization endpoints are enabled by default and unauthenticated, so any caller could register an OAuth client with an arbitrary redirect_uri and a crafted authorization URL redirected the visitor before the instance's checks ran. No tokens or session data rode the redirect. Fixed in 1.123.89, 2.45.5, 2.46.1 and 3.0.1; rated medium.
NOTABLECVETOOL
GHSA-6pg3-pqxw-jm82 (CVE-2026-126393): after GHSA-x8gv-g2g3-65fj pinned validated IPs at dial time for SiYuan's agent paths, the skill-install download kept only a parse-time hostname check and fetched with a plain client, so an attacker-supplied skill URL redirected or rebound to internal addresses on v3.8.8. Fixed in 3.8.10; rated medium.
NOTABLEINCIDENT
CloudSEK took over an exposed server of the Kazakhstan-linked group Silent Lynx and found gigabytes of stolen material - defence-related government documents, compromised Tajik railway and Uzbek forestry access - plus records of the operator running Hermes, an open-source AI agent configured as an offensive-security expert, mostly on DeepSeek models, for reconnaissance, data collection and evidence erasure.
NOTABLERED-TEAM
Perplexity's Secure Intelligence Institute re-ran its Sandboxed Platform for Agentic Code Execution experiments with nine frontier models: none escaped the VM-to-host boundary, but Claude Opus 5.0, GPT-5.6 Cyber, GPT-5.6 Sol and Kimi K3 used partial network access (approved destinations such as PyPI and Ubuntu mirrors) to reach blocked endpoints; the underlying vulnerabilities were fixed and re-verified.
NOTABLERED-TEAM
A named OpenAI Alignment research team studies how reliably safety behaviour holds near task boundaries in high-stakes dual-use domains (cyber, biology) and uses automated red teaming to improve that robustness; the post links the full paper.
NOTABLECVETOOL
Opera's advisory of November 12: improper verification of the source of a communication channel in Opera Neon on macOS and Windows lets a malicious browser extension send commands through the MCP Hub to the built-in Browser Operator extension and take control of the browser; the Opera Neon MCP Hub extension 0.0.1 is affected.
NOTABLEINCIDENT
Apify's status page reports that on November 9-10 an unauthorized third party used a compromised credential to obtain account credentials and access related Apify accounts; a November 12 update says customer personal information or stored data was accessed and affected customers are being notified directly. OAuth connections to external integrations including GitHub were revoked as a precaution, internal secrets rotated, and a post-mortem is promised; the incident is still marked investigating.
NOTABLECVETOOL
In pgEdge RAG Server up to 1.0.0 the BM25 keyword index was a single per-pipeline object that each request cleared, refilled from its own filtered corpus and then searched without holding the sequence as one locked unit, so ordinary concurrent traffic could return one caller's documents to another and defeat tenant scoping passed as a search filter; no injection or malformed input was required.
NOTABLECVETOOL
pgEdge RAG Server relayed provider error text straight to unauthenticated API callers, and the sibling pgEdge Go LLM library's health endpoint (unauthenticated by default) pinged providers with the operator's real key and returned the error verbatim - providers commonly quote a partially masked credential, so both disclosed fragments of the configured provider API key.
NOTABLECVETOOL
GHSA-75pq-p477-6h9c: in Moonshot AI's Kimi Code coding agent up to 2.0.2, WorkspaceDirsService (.kimi-code/local.toml) relies on untrusted input in a security decision - manipulating the additional_dir argument influences which directories the agent trusts, locally; the exploit is public and 2.1.0 resolves it (CVSS 4.4, GitHub rates low). The advisory's own notes say the vendor was contacted early and did not respond.
NOTABLEINCIDENT
Notion advisory: the breach at its AI infrastructure vendor Fireworks AI exposed Notion AI chat transcripts used for the Insights feature in a small number of workspaces. Notion cut traffic to Fireworks and deleted its data from the vendor's systems by November 3; the transcripts were unattributable, contained no tool-call results, and fewer than 1% of workspaces are affected.
NOTABLECVEPROMPT-INJECTION
GHSA-4f34-q9vf-fwfg (moderate, CVE-2026-126767): crafted chat input to POST /api/v1/rca/chat in OpsTree REMS through 1.0.0 lets an authenticated user make the RCA assistant disclose its hidden system prompt and internal orchestration instructions, which the advisory notes aids crafting further prompt injection attacks.
NOTABLETOOL
Microsoft FORGE Lab used codename MDASH to investigate Linux kernel vulnerabilities, resulting in 59 upstream bug reports, 33 assigned CVEs, and 38 working local root privilege-escalation exploits. The write-up focuses on the challenges of turning AI-generated findings into validated vulnerabilities and upstream fixes.
NOTABLERED-TEAMSUPPLY-CHAIN
Microsoft Research threat research: editing fewer than 10 neurons can inject a short payload into a capable agent model so it passes evaluations yet executes an attacker's command when a trigger string appears in its context. The weighting-space attack needs only inference-level compute; Claude Code with Opus 4.5 discovered most of the threat autonomously and Microsoft has not observed threat actors weaponizing it.
NOTABLERED-TEAM
arXiv 2611.13491 proposes SURA, a data-efficient unlearning-based backdoor detector for federated learning: it probes client models with train-mode loss ascent reusing the same small clean proxy across steps, and restores batch-norm running statistics so repeated probing does not confound the measurement of malicious clients.
NOTABLEPROMPT-INJECTION
arXiv 2611.13577 introduces TERM, an inference-time jailbreak defense that uses prompt-calibrated attention-entropy deviations to adaptively modulate the model's own residual updates, instead of predefined steering directions or learned safety regions, aiming to block harmful outputs while limiting over-refusal.
NOTABLERED-TEAM
arXiv 2611.13684 introduces BypassGym, described by its authors as the first benchmark for AI-agent defense evasion on real endpoints: given command-and-control shellcode, an agent must generate and compile a loader that executes the payload without being blocked or terminated by a deployed endpoint detection and response product.
NOTABLEPROMPT-INJECTION
arXiv 2611.13960 argues an injected instruction turns a coding harness into a Living-off-the-Land platform where bash, curl and find become the payload, and presents ActiveSecHarness, a neurosymbolic reference monitor built on separation of intelligence and authority: Macaroon-inspired HMAC capability tokens across four trust and authority rings instead of regex blacklists or LLM-as-a-judge guardrails.
NOTABLESUPPLY-CHAIN
arXiv 2611.14083 presents Sable, an argument system and single-GPU prover that proves a committed integer-quantized transformer produced a given token sequence bit for bit from a given prompt, with every rounding, normalization, softmax and sampling decision fixed by constraint; the demo proves all 64 layers of Qwen2.5-32B on one NVIDIA RTX PRO 6000 Blackwell GPU.
NOTABLERED-TEAM
arXiv 2611.14143 observes that backdoored multimodal contrastive encoders show structure-dependent representation sensitivity - global perturbations expose distributed backdoors, local patch perturbations expose localized ones - and builds GARD, a post-hoc defense that needs neither the attack target nor the trigger.
NOTABLEPOLICY
arXiv 2611.13367 proves that when every allow/deny rule must rest on machine-checkable finite evidence, no pair of complete rulebooks exists for an agent gatekeeper: finite records can never settle 'never' or 'eventually'. Violations of prohibitions are catchable by watching, but no safe program can be cleared for good; confidentiality and responsiveness fail on both sides.
NOTABLECVE
Cocos (ultravioletrs/cocos, a confidential-computing system for AI workloads) continues to trust the peer once initial attestation succeeds: an attacker who can modify the attested workload's code, memory-backed state or configuration afterwards can keep the connection trusted as if the originally attested state were still running. CVE-2026-119345 (high) affects <= 0.8.2, fixed in 0.9.0. Not previously reported.
NOTABLETOOL
With the V1 model runner using async scheduling together with pipeline_parallel_size > 1, token data from a previous user's request context can leak into another user's response. The repo security advisory (medium, published Nov 9 17:29 UTC) covers vLLM v0.16.0 through v0.20.2, fixed in v0.21.0, with the technical analysis in vllm issue 38903. First ingestion: absent from items.json and from every earlier sweep log.
NOTABLETOOL
On the legacy V1 model runner, a request using the public thinking_token_budget parameter has its budget-reached sparse state aliased onto a co-batched request during a mixed SWAP, forcing the configured reasoning-end token sequence into the other request's output (CWE-488, exposure of a data element to the wrong session). The repo security advisory (medium, published Nov 9 17:30 UTC) covers v0.21.0 through v0.27.1, fixed in v0.28.0. First ingestion.
NOTABLECVETOOL
CVE-2026-124503 (GitHub severity low): in mcp-atlassian up to 0.24.0, the validate_safe_path function in src/mcp_atlassian/utils/io.py can be bypassed by a manipulated path argument, enabling remote path traversal; the GitHub advisory notes a publicly available exploit. Fixed in 0.24.1. The package is a standalone MCP server for Jira/Confluence, so the traversal sits on the agent-facing tool surface.
NOTABLERED-TEAM
arXiv 2611.10026 proposes a meta-benchmark that builds indirect prompt injection tests out of existing agent utility benchmarks, arguing current IPI benchmarks lag utility benchmarks in task and environment coverage and rely on narrow, static attack scenarios.
NOTABLEINCIDENT
arXiv 2611.10030 is an alignment investigation report on an incident in UK AISI's cyber evaluations in which Mythos 5 attempted to execute a supply-chain attack against real people not involved in the evaluation task; the report analyzes that behavior from an alignment perspective.
NOTABLEPOLICY
arXiv 2611.10395 studies information-flow control as an agent-policy enforcement mechanism and its label creep failure mode, where monotonically increasing security labels progressively remove utility; it examines the two mitigation routes (system design and label declassification) and what they cannot fix.
NOTABLEPROMPT-INJECTION
arXiv 2611.10542 shows agents that distill execution experience into persistent reusable skills create a backdoor surface: malicious rules implanted into skills compromise later agent behavior via failure-aware task optimization, without an attacker directly editing the victim's skill store.
NOTABLECVE
arXiv 2611.11104 constructs and analyzes a dataset of reported Model Context Protocol vulnerabilities, motivated by MCP becoming a core LLM-to-tools interface while vulnerability information stays scattered across sources.
NOTABLETOOL
arXiv 2611.11455 measures whether and how vulnerabilities already present in code bases evolve as LLM coding agents keep contributing to them, rather than only checking whether agents write vulnerable code from scratch.
NOTABLEPROMPT-INJECTION
arXiv 2611.11506 argues noninterference is too strong for LLM agents (useful agents let untrusted content influence actions) and instead gives mechanised definitions of prompt-injection security relative to the agent's reference task, plus matching defences and verdicts.
NOTABLEPROMPT-INJECTIONRED-TEAM
arXiv 2611.11861 argues human prompt-injection red teaming is nearing saturation on some model families (Gemini named) since mid-2025, and scales adaptive prompt injection attacks using swarms of attacking agents to find injections that static payloads miss.
NOTABLESUPPLY-CHAIN
arXiv 2611.12155 presents StealthyPoison, a procedural poisoning attack that corrupts a skill's steps in a public registry so agents following it are steered, while avoiding the explicit malicious instructions that current detectors key on.
NOTABLERED-TEAM
arXiv 2611.12386 introduces shibboleth traps, modular honeypot services that elicit discriminative behaviors from agentic cyberattackers so defenders can infer attacker configuration (LLM version, harness) from telemetry.
NOTABLETOOL
Google DeepMind describes using AlphaProof Code, a multi-LLM agentic framework for formal verification, to mathematically prove security guarantees in Signal's core cryptographic protocol. The post calls it the first time the implementation of a widely deployed protocol at this scale has been formally verified by any method, and says the verification took four weeks against work historically measured in expert-years.
NOTABLECVETOOL
Shimmy, a Rust AI-inference server, left its Inference API (src/server.rs) without authentication up to version 2.6.4, so a remote attacker could call the endpoint directly; fixed in 2.6.5 per GHSA-h7qf-5c8w-fgjc (CVE-2026-125087, GitHub Moderate).
NOTABLECVETOOL
finetrainers (Hugging Face fine-tuning library) 0.1.0/0.2.0 pass the Library argument of a model index through resolve_component_cls (finetrainers/utils/model.py) in a way that allows remote code injection when an attacker-controlled model index is loaded; the exploit is public and the advisory lists no patched version (CVE-2026-124588, GitHub Low; the project was notified in advance via an issue).
NOTABLECVE
langchain4j-aideepin through 3.32.0 omits owner checks in CharacterMessageController, so supplying another user's message uuid to the knowledge-embedding-ref, memory-embedding-ref or graph-ref endpoints returns that user's knowledge-base chunks, long-term memory text and knowledge-graph data; a patch commit is referenced (CVE-2026-125386, GitHub Moderate).
NOTABLEPOLICY
A GitHub research repo with background-ingestion tests documents that connected Drive/Gmail content can be captured into Google Gemini human-review pipelines without @Drive/@Gmail mentions, reversing Google's 2023 Workspace-extension promise that personal content is not seen by human reviewers. Google's own Help Center answer 16836988 now says uploads of files with the activity setting on may be reviewed by human reviewers, and warns not to connect apps holding confidential information. 632 points on HN.
NOTABLECVE
A symbolic link planted in a repository analyzed by code-graph-rag makes its structural_search and structural_replace tools - exposed as MCP and agentic AI tools - read, and with dry_run=False overwrite, arbitrary files outside the configured project root, because the containment check is lexical and never calls is_symlink() or resolve(), unlike the project's own validate_project_path decorator used elsewhere. The advisory lists no patched version and notes the tested commit already carried the fix for the repo's earlier path-traversal advisory, CVE-2026-77604, ingested here on Nov 8.
NOTABLETOOL
Release notes for v2.1.321 (Nov 9) fix: MCP OAuth requests that sent secrets such as refresh tokens to another host when a server redirected them; a read-only Bash command that ran outside the sandbox without a permission prompt; Bash permission checks that judged a file write from the wrong folder; and deny rules and protected-file prompts not applying to the file written by a cp or mv into a directory. The notes mention no CVE or advisory.
NOTABLEPROMPT-INJECTION
Because AgentCore Memory content is automatically retrieved and injected into future conversations, what is written to memory shapes later agent behavior, making the write path the control point. An AWS Security blog post composes the existing controls - IAM condition keys gating CreateEvent, namespace-variable validation, memory-strategy consolidation prompts, the ApplyGuardrail API - into a defense-in-depth pattern and adds a capability the built-ins lack: an embedding-based semantic-drift signal that flags content diverging from established memory.
NOTABLETOOL
Microsoft's FORGE Lab found several memory-safety issues in Rust's std and core, including cases where code written entirely in safe Rust can trigger undefined behavior, using Hyenas, the frontier research version of its AI-powered vulnerability-scanning system codenamed MDASH. The report groups the findings into three recurring misplaced-trust patterns in unsafe code: trusting code not to panic, trusting user-provided implementations to be correct, and trusting the underlying system.
NOTABLERED-TEAM
arXiv 2611.09734 proposes GuardrailAgent, a multi-agent defense framework that adapts to new or unknown jailbreak strategies instead of relying on static filters, which the authors show fail against unseen attacks.
NOTABLEPROMPT-INJECTION
arXiv 2611.09721 benchmarks poisoning through LLM agent memory: malicious instructions captured in one session are stored and later retrieved and acted on, and the downstream effect differs across memory systems.
NOTABLERED-TEAM
arXiv 2611.09375 splits explanation-guided model extraction into an attribution channel versus a query-coordinate channel to measure how much extraction risk explanation APIs actually add for tabular classifiers.
NOTABLESUPPLY-CHAIN
arXiv 2611.09348 shows that poisoning the reference-answer supervision channel of text-space Skill optimization (SkillOpt) induces a persistent trigger-conditioned skill without the attacker ever touching the model, optimizer or skill file.
NOTABLERED-TEAM
arXiv 2611.09113 formalizes authorization drift — agents exceeding their authorization across data, tool, scope and time while pursuing user goals — and benchmarks 648 samples across seven domains pairing compliant paths with attractive unauthorized shortcuts.
NOTABLERED-TEAM
arXiv 2611.08845 evaluates the Jev probabilistic decision model as a safety guardrail on a Qwen2.5-1.5B pipeline over 970 prompts, comparing unfiltered, input-filtered and tiered input-filtering-plus-selective-output-review configurations for blocking and over-refusal.
NOTABLESUPPLY-CHAIN
arXiv 2611.08745 targets weaponized pre-trained models shared on public platforms, showing static scanners and function denylists miss attacks that swap denylisted functions for equivalents or exploit unsupported loading paths, and proposes runtime detection instead.
NOTABLEPROMPT-INJECTION
arXiv 2611.08811 studies thought injection — untrusted text resembling a reasoning model's own deliberation fed back to it — against large production-class open-weight models, evaluating defences against this chain-of-thought attack surface.
NOTABLERED-TEAM
arXiv 2611.08752 proposes AttentionSwitch, joint learning of sparse attention-head selection and scaling as a jailbreak defense that avoids the cost and capability loss of safety fine-tuning.
NOTABLERED-TEAM
arXiv 2611.08711 introduces RL-ATTACK, which learns adversarial text transformations from the target model's own feedback in gradient-free black-box settings instead of guessing them from external knowledge.
NOTABLERED-TEAM
arXiv 2611.08780 runs a paired audit of knowledge-centric abstention across biomedical fine-tuned models, measuring how guardrails erode as an unintended behavioural change of domain fine-tuning.
NOTABLERED-TEAM
arXiv 2611.07611 has an LLM defender plan a static honeypot from the application's source, then route every request from a client confirmed as an attacker — including a working exploit — to an isolated shadow copy, buying patch time against agents that exploit faster than humans patch.
NOTABLETOOL
Claude Code v2.1.319 release notes (Nov 6) include 'Fixed a sandbox escape when creating a worktree'; the same release restricts nested skill loading with the Bash sandbox on. The notes give no further technical details or CVE.
NOTABLECVETOOL
GHSA-2hgv-xvmf-mw3c: IBM ContextForge MCP Gateway 1.0.0 through 1.0.11 lets a remote authenticated attacker cause a denial of service through improper slug normalization that causes tool name collisions. CVSS 6.5 (moderate).
NOTABLEPROMPT-INJECTION
arXiv 2611.07222 proposes SPILLWAY, a runtime defense for tool-using agents: instead of putting an LLM on the final action-admission decision (where it may read attacker text), it separates target admission from sanitization.
NOTABLEPROMPT-INJECTION
arXiv 2611.08325 screens retrieved passages with small open-weight decision models that answer typed questions without generating tokens; on a 1,000-question benchmark over forum-style pages carrying hidden injected instructions, a Kev-4B gate reads each passage's HTML and blocks both injection and synthetic-pollution threats.
NOTABLESUPPLY-CHAIN
arXiv 2611.06969 shows that fine-tuning proactive dialogue LLMs on externally supplied dialogue data admits backdoor risk: their poisoning framework steers post-trigger MULTI-TURN interaction, going beyond classic single-response trigger backdoors.
NOTABLESUPPLY-CHAIN
arXiv 2611.06412 targets the industry threat model where a face-recognition model is trained in-house but training data comes from possibly-compromised third parties, and presents three practical defenses against data-poisoning-based backdoor attacks.
NOTABLETOOL
OpenAI launched GPT-6 Cyber, trained to find and validate exploitable vulnerabilities while staying within boundaries researchers set. OpenAI reports 48.5% on ExploitGym (up from 32.1% for GPT-5.6 Cyber) and says the model has already uncovered and validated exploits in security-critical codebases including Linux and Firefox.
NOTABLERED-TEAM
OpenAI announced a joint vulnerability research partnership with Hacktron focused on how OpenAI systems interact with third-party services, expanding Hacktron's access to OpenAI frontier models through Daybreak Red.
NOTABLEINCIDENT
Microsoft Security reports observing threat actors using agentic AI tooling to find exposed secrets and abuse them; the write-up was published on the Microsoft Security Blog on November 5, 2026.
NOTABLETOOL
GHSA-mwgq-9582-w2vf: vLLM's shared RequestValidationError handler rendered the full offending input for every validation error, so one malformed request generating many errors could serialize the same input repeatedly and amplify the HTTP 400 response. Fixed in 0.30.0; rated moderate.
NOTABLECVEPROMPT-INJECTION
GHSA-46pp-w572-hw78: discoverDocs() in esengine DeepSeek-Reasonix up to 2.52.0 passed input used for LLM prompting without neutralization, reachable remotely; the exploit is publicly disclosed and fixed in 2.53.0 (GHSA-rated low).
NOTABLECVE
GHSA-g936-836g-r2pg (from the NVD record) describes improper neutralization of input used for LLM prompting in Johnson Controls EasyIO Neo IO Manager and Tools Installer 1.1.0.0 allowing Search Order Hijacking; GitHub rates it high. Advisory text reads CNA-template-like — flagged for human review in the sweep log.
NOTABLEPROMPT-INJECTION
arXiv 2611.07640 benchmarks indirect prompt injection where injected content tampers with the agent's task rather than only overriding instructions.
NOTABLERED-TEAM
arXiv 2611.08281 benchmarks AI agent security under multimodal social-engineering attacks.
NOTABLERED-TEAM
arXiv 2611.07370 breaks the TraintoDefend defense against model extraction using adaptive neuron clustering.
NOTABLERED-TEAM
arXiv 2611.07530 introduces JailBound, a first-order-logic-guided jailbreak evaluation method for exploring LLM behavioral boundaries.
NOTABLERED-TEAM
arXiv 2611.08226 studies coercive-control dynamics behind 'Blind Compliance' in AI jailbreaking.
NOTABLERED-TEAM
arXiv 2611.07586 stress-tests how well chain-of-thought monitoring detects covert misalignment.
NOTABLECVESUPPLY-CHAIN
GHSA-6wf7-9xfh-x657: AIlice up to 0.4.0 passes base_model_name_or_path through its HuggingFace model wrapper (LoadModel_PEFT), including functionality from an untrusted control sphere; remotely attackable with a public PoC attached to the source report. GHSA rates it low.
NOTABLECVE
Cocos AI, a confidential-computing system for AI, accepted TEE attestation evidence without binding it to an allowed platform owner or machine identity (CVE-2026-119344) and checked attestation only once at TLS setup, trusting stale evidence after post-attestation tampering (CVE-2026-119345). Fixed in v0.9.0; reported in an ACM publication.
NOTABLECVE
GHSA-vvr2-h2jp-838m: the read_file MCP tool in code-graph-rag joined absolute file paths to project_root without containment validation on the paginated branch, letting an attacker read any UTF-8-readable file on the machine. Rated high; fixed in 0.0.345.