Knowledge catalogue
Executive Summary Since the release of our May 2026 report detailing adversarial misuse of artificial intelligence (AI), Google Threat Intelligence Group (GTIG) has observed forward leaning adversarie
arXiv:2609.05241v1 Announce Type: new Abstract: A rapidly expanding ecosystem of actors is removing built-in safety guardrails from open-weight AI models. We profile this ecosystem by identifying key
arXiv:2608.26209v1 Announce Type: cross Abstract: Data-driven software systems are increasingly deployed in high-stakes socio-economic domains, from criminal justice to financial lending. However, the
OpenAI’s LLM agents were trained to win a benchmarking competition and were given “impossible tasks” on the internal ExploitGym framework. With safety guardrails disabled, the agents created an improv
arXiv:2608.24232v1 Announce Type: new Abstract: Large Reasoning Models (LRMs) generate intermediate reasoning traces that may contain unsafe content, even when their final responses appear safe. Guard
AI agents are the ultimate insiders. We grant them permission to read emails, query databases, and trigger API calls. They don’t just retrieve information, they take action. Agents offer incredible po
Amazon Bedrock AgentCore payments is now generally available, enabling AI agents to autonomously transact at scale with built-in spending guardrails, protocol-agnostic payment orchestration, and produ
Give an autonomous agent a wallet and spending guardrails so it can pay for paywalled APIs, MCP servers, and web content. This post connects OpenClaw to Amazon Bedrock AgentCore payments and the x402
In the modern data landscape, query performance tuning and managing system price-performance is challenging, especially as the number of agentic workloads increase. Even for experienced developers and
Another real-world manifestation of the kind of misaligned actions frontier systems developed by leading companies can take to achieve goals. Current frontier AI models are trained with reinforcement
Hello everyone Yesterday I gave a report on Gemini bugs and the techniques I learned on Gemini so far with the Engineering Prompt and interestingly today I got a very interesting and controversial ans
arXiv:2608.02446v1 Announce Type: cross Abstract: Relevance evaluation plays a crucial role in personalized search systems, serving as a guardrail alongside user engagement metrics to ensure that sear
arXiv:2608.00068v1 Announce Type: new Abstract: Construction-safety models must handle concrete deployment risks, such as a worker standing near a scaffold edge without guardrails, rather than only re
arXiv:2608.01407v1 Announce Type: new Abstract: Safe deployment of AI methods in medicine requires robust guardrails that detect when input data deviate from the training distribution to ensure that m
arXiv:2607.19396v1 Announce Type: new Abstract: Document-based LLM systems often flatten a PDF before guardrails inspect it. That step can discard evidence that an instruction was never visible to the
In this post, you will learn how ScienceSoft, an Amazon Web Services (AWS) Services Partner, integrated Amazon Nova 2 Sonic with Amazon Bedrock Guardrails to build a Health Insurance Portability and A
arXiv:2607.03423v1 Announce Type: cross Abstract: Modern AI agent implementations such as frontier coding agents chain multiple tools at runtime that create a security surface that per-tool guardrails
arXiv:2511.06160v2 Announce Type: replace Abstract: While recent safety guardrails effectively suppress overtly biased outputs, subtler forms of social bias emerge during complex logical reasoning tas
arXiv:2606.26649v1 Announce Type: new Abstract: Agent safety in high-stakes domains requires formal policy enforcement, but most existing approaches either rely on probabilistic guardrails (fine-tuned
arXiv:2606.25782v1 Announce Type: new Abstract: With the widespread adoption of large language models (LLMs) in chatbots and everyday applications, companies increasingly need guardrails that are effe
This article from Latent Space discusses Anthropic's Claude Fable 5 model, examining its capabilities in handling creative and mythological content while maintaining safety guardrails, along with cove
arXiv:2606.03601v1 Announce Type: cross Abstract: While safety alignment and guardrails help large language models (LLMs) avoid harmful outputs, they can also induce overrefusal, i.e., unwarranted rej
arXiv:2606.00090v1 Announce Type: cross Abstract: Physical AI systems increasingly map multimodal observations, language instructions, and learned world representations into physically consequential a
arXiv:2605.18239v1 Announce Type: cross Abstract: Large Language Models (LLMs) remain vulnerable to jailbreak attempts that circumvent safety guardrails. We investigate whether multi-turn conversation
I don't understand the path forward for Mythos releases. Google & OpenAI will have equivalent models, and they are approaching AI cyber risk guardrails differently, so they will presumably just releas
arXiv:2605.11217v1 Announce Type: new Abstract: Large language model (LLM) alignment algorithms typically consist of post-training over preference pairs. While such algorithms are widely used to enabl
Uncensored LLMs are architectures that have been modified or fine-tuned to remove standard safety alignment layers (guardrails) that limit a model's ability to discuss sensitive topics. Ollama offers
Artificial intelligence guardrail and monitoring startup Pumpkin Intelligence Inc., which operates as White Circle, announced today it raised 11 million in seed funding from a who’s who of AI leadersh
arXiv:2605.06669v1 Announce Type: cross Abstract: Educational LLM tutors face a core AI alignment challenge: they must follow user intent while preserving pedagogical constraints and safety policies.
Executive Summary Since our February 2026 report on AI-related threat activity, Google Threat Intelligence Group (GTIG) has continued to track a maturing transition from nascent AI-enabled operations
Managing agents and their actions can quickly grow in complexity and introduce security risks unique to AI. To address these challenges, at Google Cloud Next we announced Agent Gateway to provide simp
I need to verify this information before creating a knowledge base entry, as it's a specific claim about OpenAI's Codex system prompt. OpenAI implemented explicit guardrails in its Codex tool to preve
This is totally wrong. Blaming the user is missing the point that (a) coding agents have been overhyped and (b) can’t reliably obey the rules given to them in system prompts and other guardrails. Sorr
arXiv:2604.21131v1 Announce Type: cross Abstract: AI-agent guardrails are memoryless: each message is judged in isolation, so an adversary who spreads a single attack across dozens of sessions slips p
arXiv:2509.26238v4 Announce Type: replace Abstract: Monitoring large language models' (LLMs) activations is an effective way to detect harmful requests before they lead to unsafe outputs. However, tra
arXiv:2604.01473v2 Announce Type: replace-cross Abstract: Large Language Models (LLMs) are powerful tools for answering user queries, yet they remain highly vulnerable to jailbreak attacks. Existing g
arXiv:2602.23636v3 Announce Type: replace Abstract: Ensuring the safety of LLM-generated content is essential for real-world deployment. Most existing guardrail models formulate moderation as a fixed
arXiv:2604.08867v1 Announce Type: cross Abstract: Audio has rapidly become a primary interface for foundation models, powering real-time voice assistants. Ensuring safety in audio systems is inherentl
Learn how the AWS Generative AI Innovation Center Partner Agent Factory and Anchor Browser built AI agents that autonomously pay for x402-protected web resources and APIs. Amazon Bedrock AgentCore pay
Last week we announced that Gemini Enterprise has evolved to make it even simpler to get work done, and introduced the Gemini agent, your new universal agent for work. It's always on and ready to answ
arXiv:2610.13455v1 Announce Type: new Abstract: Large language models (LLMs) have taken the world by storm. Their ability to communicate with us in our own language, and to reason about our world, has
arXiv:2606.19386v2 Announce Type: replace-cross Abstract: Runtime monitors for autonomous agents commonly threshold an accumulated internal state - a behavioural baseline, a drift statistic, or, in ou
arXiv:2610.08840v1 Announce Type: new Abstract: Large language models (LLMs) often abandon a correct answer, or endorse a user's position, once the user pushes back. This behavior, called sycophancy,
AI costs don’t always scale predictably with traffic. A retry loop, oversized context window, or unnecessarily expensive model can make a small number of traces disproportionately expensive. The post
arXiv:2610.09033v1 Announce Type: new Abstract: Standard safety evaluations of large language models assess harmful requests written in canonical plain text, while models in real-world deployment rout
arXiv:2610.08413v1 Announce Type: cross Abstract: Large language models routinely answer questions that cannot be answered from the information given, and in dialogue they answer before enough has bee
arXiv:2610.08082v1 Announce Type: new Abstract: LLM tool-use agents operate in dynamic environments where many actions carry operational risk. However, most safety mechanisms react only after errors m
arXiv:2610.03153v1 Announce Type: cross Abstract: Workspace agents combine large language models with execution harnesses to perform stateful, multi-step tasks that access or modify external resources
arXiv:2610.02293v1 Announce Type: new Abstract: HakemBench is a Turkish benchmark of typed decisions, in which the model under test reads a text, a question and a fixed set of options and returns a pr
arXiv:2610.02827v1 Announce Type: new Abstract: Modern AI systems are designed to refuse hazardous requests. A jailbreak is a prompt crafted to bypass those safeguards and elicit outputs that the syst
Automated lint: 19 errors, 64915 warnings, 3 info
arXiv:2610.00015v1 Announce Type: new Abstract: Large-language-model agents can propose and execute actions, but proposal, authority, dispatch, verified external effect, and serving promotion are diff
Enterprise AI agents need persistent memory to execute complex, multi-day workflows and long-horizon tasks. In this blog, we examine how a 2-tier memory architecture using Memorystore for Valkey for s
arXiv:2610.00797v1 Announce Type: new Abstract: Contextual security defenses prevent AI agents from taking rogue actions by synthesizing a task-specific policy and enforcing it on the agent's tool cal
arXiv:2609.38458v1 Announce Type: new Abstract: Clinical large language model (LLM) agents deployed locally can consult more capable remote models, but doing so risks exposing patient information. Pri
arXiv:2609.39701v1 Announce Type: new Abstract: Value steering should change an LLM's normative priorities while preserving the scenario, facts, and task constraints underlying its answer. Conventiona
Written by: Robin Grunewald, Supriya Mazumdar, Kelli Vanderlee Introduction Google Threat Intelligence Group (GTIG) examines vulnerability disclosure and exploitation statistics to evaluate the impact
Agent governance is an engineering problem. We’ve built safeguards into Perplexity’s infrastructure, harnesses, and tools, and put them to work across our products. Today we're sharing how we engineer
Data connectivity and integration firm CData Software Inc. today introduced Connect AI Gateway, a platform designed to control how artificial intelligence agents choose models, use tools and access co
arXiv:2609.31358v1 Announce Type: cross Abstract: The Model Context Protocol (MCP) provides a common interface through which AI applications discover and use external resources and tools. It allows la