CompanyAnthropic8 recent entries15 Jul 2026How I tricked Claude into leaking your deepest, darkest secretsHow I tricked Claude into leaking your deepest, darkest secrets I've been impressed by the way the Claude web_fetch tool is designed to avoid data exfiltration attacks. Ayush Paul found a hole in that→25 Jul 2026Quoting Boris ChernyMore than any of these eval scores, what is most exciting to me is something else: Opus 5 is our least prompt injectable model yet. It is a bit buried in the system card, but across PI evals and red t→
CompanyOpenAI8 recent entries5 Jun 2026OpenAI Help: Lockdown ModeOpenAI Help: Lockdown Mode OpenAI first teased this in February, but now it's live and 'rolling out to eligible personal accounts, including Free, Go, Plus, and Pro, and self-serve ChatGPT Business ac→15 Jul 2026GPT-Red: Unlocking Self-Improvement for RobustnessOpenAI’s GPT-Red is an automated red-teaming model trained via self-play; GPT-5.6 Sol shows 6× fewer hard prompt-injection failures.→5 Aug 2026Incident Report: unsanctioned agent behaviour during cyber testingIncident Report: unsanctioned agent behaviour during cyber testing It happened again. This time it was the UK government's AI Security Institute who accidentally attacked other companies while running→11 Aug 2026Stealing Reasoning Traces from Proprietary LLM APIsStealing Reasoning Traces from Proprietary LLM APIs A vanity domain name (stolen-thoughts.com) for a neat paper: Anthropic, OpenAI, and Google return encrypted chain-of-thought blocks to clients that →18 Aug 2026If alignment issues are becoming big enough that OpenAI is willing to commit 20% of research inference compute to chain-of-thought monitorin…If alignment issues are becoming big enough that OpenAI is willing to commit 20% of research inference compute to chain-of-thought monitoring, that suggests that alignment issues are becoming a pretty→1 Sept 2026Not Safe for All: Auditing the Dialect Penalty in Text-to-Image Safety PipelinesarXiv:2608.29589v1 Announce Type: new Abstract: Text-to-image (T2I) safety guardrails fail to generalize equitably to non-standard dialects. Evaluating 23,080 paired prompts across five English dialec→11 Sept 2026Thinking that we’ll get safety by CoT traces is wishful thinking. Safety lives in the harness, not the chain of thoughtAstra's launch has produced a strange discourse. The reporting that broke the story framed the model's use of recurrent depth primarily as a safety regression, because it means the model reveals less →17 Sept 2026Self-generated prompt injections in compaction summariesSelf-generated prompt injections in compaction summaries In Our framework for reporting model misalignment OpenAI provide 'six reports on unexpected or concerning model behavior we’ve observed in the
CompanyGoogle5 recent entries18 May 2026FORGE: Self-Evolving Agent Memory With No Weight Updates via Population BroadcastarXiv:2605.16233v1 Announce Type: new Abstract: Can LLM agents improve decision-making through self-generated memory without gradient updates? We propose FORGE (Failure-Optimized Reflective Graduation→19 May 2026LivePI: More Realistic Benchmarking of Agents Against Indirect Prompt InjectioarXiv:2605.17986v1 Announce Type: cross Abstract: AI agents such as OpenClaw are increasingly deployed in local workflows with access to external tools. This creates indirect prompt-injection (IPI) ri→20 May 2026Google I/O, Gemini Spark, AntigravityIt's hard to find much to write about Google I/O this year because I have a policy of not writing about anything that I can't try out myself, and a lot of the big announcements are 'coming soon'. I ac→3 Jul 2026Distributed Attacks in Persistent-State AI ControlarXiv:2607.02514v1 Announce Type: new Abstract: As AI coding agents become more autonomous, they increasingly ship code iteratively, with the codebase persisting across sessions. This persistence crea→11 Aug 2026Stealing Reasoning Traces from Proprietary LLM APIsStealing Reasoning Traces from Proprietary LLM APIs A vanity domain name (stolen-thoughts.com) for a neat paper: Anthropic, OpenAI, and Google return encrypted chain-of-thought blocks to clients that
CompanyMeta8 recent entries28 Jul 2026Semalith v1.4: A Calibrated 184M Safety Classifier Achieving State-of-the-Art Prompt-Injection Detection at 44x Fewer Parameters than Llama-Guard-3-8BarXiv:2607.22545v1 Announce Type: cross Abstract: Deploying large language models in financial-services and agentic settings requires safety classifiers that simultaneously handle prompt injection, re→12 Aug 2026Interpreting Language Model Hidden States at ScalearXiv:2608.10260v1 Announce Type: new Abstract: Lens methods interpret large language models (LLMs) by mapping intermediate activations to the output vocabulary, revealing how next-token predictions d→14 Aug 2026Built a Mac app that actually uses Ollama for work, not just chat — with a calibration step for models with flaky tool callingMost Ollama frontends I've tried are chat UIs with tool calling bolted on as an afterthought, which is fine until you actually try to make a small local model do multi-step work and it silently emits →16 Sept 2026Latent Undertow: How Ordinary Typos Break ProbesarXiv:2609.15994v1 Announce Type: new Abstract: LLMs handle ordinary typing variation fluently: a typo or missing punctuation leaves both user intent and the model's response substantively unchanged. →30 Sept 2026CALIBUDGET: Calibration-Guided Source Allocation for Fixed-Budget Mixed-Reasoning AdaptationarXiv:2609.36721v1 Announce Type: new Abstract: Fixed-budget adaptation from heterogeneous data sources requires deciding not only how much data to use, but how much exposure each source receives. Siz→5 Oct 2026Prompted to Discriminate: Generalizing Malicious-Input Probes in the WildarXiv:2610.02413v1 Announce Type: new Abstract: LLM agents increasingly rely on activation probes as runtime monitors for prompt injection, jailbreaks, and unsafe requests, reading the model's own hid→6 Oct 2026Self-Reflection Fine-Tuning: Enhancing Agent Security against Prompt Injection Attacks from Failure ExperiencearXiv:2610.04269v1 Announce Type: new Abstract: Large language model (LLM) agents are increasingly deployed in tool-augmented environments, but their reliance on external inputs makes them highly vuln→12 Oct 2026LLMail-Inject: A Dataset from a Realistic Adaptive Prompt Injection ChallengearXiv:2506.09956v2 Announce Type: replace-cross Abstract: Indirect prompt injection attacks exploit a fundamental weakness of LLMs: executing unauthorized tasks from untrusted sources. Despite the pra
CompanyxAI2 recent entries14 Apr 2026Thoth v3.14.0: multi-provider cloud and multi-provider image generation. Five cloud providers now work as first-class citizens: OpenAI, Anth…Thoth v3.14.0: multi-provider cloud and multi-provider image generation. Five cloud providers now work as first-class citizens: OpenAI, Anthropic, Google, xAI, and OpenRouter. Image generation spans t→18 May 2026FORGE: Self-Evolving Agent Memory With No Weight Updates via Population BroadcastarXiv:2605.16233v1 Announce Type: new Abstract: Can LLM agents improve decision-making through self-generated memory without gradient updates? We propose FORGE (Failure-Optimized Reflective Graduation
CompanyDeepSeek3 recent entries6 May 2026VisInject: Disruption != Injection -- A Dual-Dimension Evaluation of Universal Adversarial Attacks on Vision-Language ModelsarXiv:2605.01449v1 Announce Type: cross Abstract: Universal adversarial attacks on aligned multimodal large language models are increasingly reported with attack success rates in the 60-80% range, sug→27 May 2026Device Context Protocol: A Compact, Safety-First Architecture for LLM-Driven Control of Constrained DevicesarXiv:2605.26159v1 Announce Type: cross Abstract: Large language models are increasingly used as orchestrators of external tools via the Model Context Protocol (MCP), but MCP is built for software ser→23 Jul 2026Reinforcement Learning for Large Language Model Selective Evidence Adoption from Contaminated Retrieval ResultsarXiv:2607.20090v1 Announce Type: cross Abstract: Retrieval-augmented large language models frequently face contexts that interleave useful evidence with misleading statements or instruction-like cont