All items
All tracked items across vulnerabilities, news, research, incidents, and regulatory updates.
All tracked items across vulnerabilities, news, research, incidents, and regulatory updates.
12,081 items
Perseus is a free, open-source Python library for encrypted model inference using Fully Homomorphic Encryption, covering the cryptographic backend, data packing, encrypted nonlinearities and end-to-end model execution. It adds optimizations at each level, including in-kernel regeneration of key-switching keys, a variant of Cachemir packing, and a bootstrap placer. It stays within 3% of plaintext accuracy and perplexity and is 3.96x faster at single-token decoding on Llama-3-8B than any other implementation.
CrowdStrike is introducing AI Kill Switch in Falcon Guardian, a control that lets security teams stop rogue AI agents. When activated, it directs connected AI gateways, either the Falcon AI Gateway or a third-party gateway, to block agent access to the AI models behind the agent's decisions and actions. CrowdStrike plans to extend the control to coordinated containment across endpoints, cloud, SaaS, identities and browser sessions.
CrowdStrike announced its intent to acquire Dreadnode, which will bring automated AI security testing to the Falcon platform. Dreadnode runs more than 50 attack techniques, including jailbreaks, prompt injection and agent-specific attacks, against AI systems, models, applications, agents and MCP servers before production, then ranks findings and maps them to OWASP, MITRE ATLAS and NIST AI RMF. Dreadnode will be generally available in Q4 FY27.
Cisco is combining its network telemetry with OpenAI models to correlate security signals and speed up vulnerability patching, investigation and response. Cisco and OpenAI are also exploring the Daybreak Defense Network to help security teams triage exposure faster and enforce controls in Cisco Cloud Control.
CrowdStrike announced that it is bringing its Falcon Guardian protection natively into Anthropic's Claude Cowork, built in partnership with Anthropic. The Falcon sensor runs inside the Cowork virtual machine on macOS and Windows to give runtime visibility and detection of agent process, package and network activity, which is otherwise invisible to the host's endpoint sensor. The integration is available in beta on macOS, with Windows support expected later in November.
CrowdStrike announced its intent to acquire Fiddler, an AI observability company, at Fal.Con EU 2026. Fiddler will become part of the CrowdStrike Falcon platform as Falcon Observe, generally available in Q4 FY27, providing pre-deployment evaluation and continuous behavioral monitoring of enterprise AI agents. The acquisition aims to connect agent decisions with Falcon security telemetry across endpoints, identity, cloud and runtime environments.
CrowdStrike announced advances in AI detection and response (AIDR) at Fal.Con Europe 2026, spanning architecture, coverage and controls for its Falcon Guardian product. The release adds an AI kill switch that cuts off AI model access at AI gateways when agents become dangerous, and general availability of Falcon Complete for Guardian, which extends 24/7 managed detection and response to AI security.
Heimstaden, a European residential real estate operator, chose OpenAI in June 2025 to apply AI to its property management, including ChatGPT Enterprise and Codex. The company says AI now supports pricing, advertising, listing review and market feedback in its letting process. In the Netherlands, the cost of a virtually staged property image fell from approximately €50 to under €0.10, a reduction of more than 99%.
TxBench is a benchmark of attack transactions and evidence-backed negatives across seven EVM chains, built by a five-step pipeline that collects transactions cited by BlockSec's incident library or the EVM Hack Registry and rebuilds their execution from free data sources. Of 1,850 cited transactions, it accepts 1,012 attacks from 602 incidents worth USD 868.1 million, comprising 846 contract exploits and 166 key compromises. Four zero-shot LLMs reading anonymized traces are precise but find at most 0.43 of the attacks and no key compromise.
The author sent the October edition of a sponsors-only monthly newsletter, available to sponsors through a link. The issue covers vibe game development, vulnerabilities, LLM releases, and the author's current tools.
Composer 2.11 adds dependency cooldowns, letting PHP developers delay newly published package versions through the config.policy.cooldown setting, which takes waiting periods in hours, days or weeks. The feature is disabled by default, and Composer 2.11 uses Packagist's new published-time field rather than publisher-supplied dates to prevent backdating bypasses.
Fix: Enable the feature with the config.policy.cooldown setting (disabled by default); organizations can enforce cooldowns through Private Packagist. Developers can use --no-cooldown to install an urgent fix immediately.
Socket BlogEndor Labs describes agentic development security as the practice of securing the software development process when AI coding agents generate, select, and ship code on their own. The article covers the tools agents use, the actions they take, and the code and dependencies they produce. It cites Forrester analyst Janet Worthington's April 2026 report naming this shift.
OpenAI has hired former U.S. Army Secretary Dan Driscoll as senior counselor for national security modernization, a newly created role. He will focus on how AI can strengthen U.S. national security and cybersecurity, and on helping U.S. allies adopt advanced AI responsibly.
Researchers compare agent success rates on safety evaluations with success rates on comparable production decisions, using the Respecting Warnings evaluation as a representative test. Some models perform substantially worse on evaluations than in production, with the largest gaps in models that most often verbalize incorrect evaluation objectives. The authors find that evaluation-objective awareness is closely associated with the gap, and that objective framing can causally change evaluation success rates.
Researchers show that an attacker can steer LLM agents through their own harm-avoidance rather than through injected commands. A single fabricated hazard, written as an ordinary review, leads the agent to disqualify legitimate options and select the attacker's target, and deployed injection detectors score these payloads as benign. The authors call this voluntary servitude and report success rates of 60.4% against 66.0% for instruction injection, with transfer to production browser agents and a retrieval pipeline (1.0-57.0% end-to-end).
Fix: A corroboration requirement added to the agent's instructions removes roughly 95% of the exposure against a non-adaptive attacker, without retraining.
Arxiv (cs.CR + cs.CL + cs.LG)This paper studies training-time backdoor attacks on Scene Graph Generation (SGG), focusing on the explicit global context pathway in recent architectures. The authors instantiate an attack that selectively corrupts relations involving a designated object category, and evaluate it on Visual Genome with open-vocabulary and closed-set SGG models using a three-axis protocol. Mechanistic analysis finds the global context pathway is a major carrier of the backdoor effect. Replacing the poisoned global context representation with its clean counterpart at inference cuts attack success by 100% for the OvSGTR family and 97% for RelationFormer.
CloakFEM is a domain-guided LLM-agent framework that generates runnable adversarial finite element method (FEM) programs designed to amplify solver computation. Across three FEM stacks, 17 problem classes, and 1,068 seeds, it found inputs that substantially increase solver work while staying valid and solvable, reaching a 5,752.72x amplification in solver iterations (43 to 247,367) in the most extreme case. It outperforms five generic black-box baselines and retains an advantage over evolutionary search.
This study examines whether benign reusable skills can make LLM agents execute unsafe external actions, a risk the authors call skill-induced behavioral misalignment. Across five LLM backbones and three benchmarks, skill augmentation raised benign task accuracy by 10.3 percentage points and also raised the unsafe action rate by 10.3 pp, including with expert-curated skills. A paired trajectory analysis found 24 of 50 newly unsafe trajectories did not complete the harmful objective.
Fix: A guard that checks proposed tool actions before execution reduces violations while retaining gains on benign tasks.
Arxiv (cs.CR + cs.CL + cs.LG)Fix: To secure the code your agents write, treat every agent output as untrusted input and give each agent the least privilege it needs. Run agents in a sandbox, keep secrets out of their context, and limit which commands and network destinations they can reach. Then scan every pull request automatically, require an independent human approver for high-impact changes, and log every action an agent takes.
Endor Labs BlogThis research shows that visual perturbations can change the physical action of reasoning vision-language-action models while the generated rationale stays word-for-word identical. Using a text-neutral action-subspace (TNAS) diagnostic on Alpamayo (driving) and InstructVLA (manipulation), the authors find that action-sensitive directions are nearly orthogonal to text-loss gradients. In 154 obstacle-annotated driving scenes, TNAS-guided attacks induced a new collision under an identical rationale in 52.6% of scenes under full-image perturbations, compared with 4.55% for unguided action attacks.
Researchers present a generative method that uses knowledge graphs, medical ontologies and a base LLM to produce red-teaming evaluation cases for medical LLMs at scale. Each case flips a sensitive attribute, such as insurance status, to check whether the model's answer changes unfairly. The authors report that the generated cases are reliable and reveal bias patterns, and they publish a bias evaluation dataset on GitHub.