Thinking Machines Lab releases open-weight Inkling 2 Mini model that runs on a laptop
Thinking Machines Lab has released Inkling 2 Mini, an open-weight model built on a recurrent Transformer architecture…
The Safety and Alignment topic covers research and governance to keep AI controllable, trustworthy, and hard to misuse: jailbreaks and defenses, model behavior evaluation, red teaming, and policy frameworks. For researchers and compliance professionals following the latest progress and reusable evaluation methods.
Making AI controllable, trustworthy, and resistant to misuse: jailbreaks and defenses, model behavior research, safety evaluation, and governance frameworks.
Thinking Machines Lab has released Inkling 2 Mini, an open-weight model built on a recurrent Transformer architecture…
The MiniMax research team used its internal research model MIM to conduct automated cryptanalysis of modern…
Anthropic CEO Dario Amodei issued a statement on the company's IPO, saying it puts the company on firmer financial…
The Verge reports that Anthropic has completed its IPO, raising $81.3 billion and surpassing SpaceX to become the…
The OpenHands team details how it hands most CVE remediation work to an agent pipeline: agents select fixes, open PRs,…
Google DeepMind releases Gemini 4 Neon, focused on frontier intelligence and fast end-to-end task completion across…
Internal chats disclosed by Reuters show that Meta employees testing the Muse agent found it appeared to access…
Researchers at Stanford HAI and RegLab released the paper Detection Objection and the PhantomDocket benchmark,…
Independent evaluator Vals AI tested models from eight vendors and found that mainstream large language models have…
Wiz Research, along with Ikotas Labs, Rapid7 and other teams, independently discovered two critical remote code…
OpenAI published a long-form article detailing how the Codex sandbox runs commands within users' existing environments…
OpenAI published a long-form article describing its defense-in-depth system for managing frontier AI risks across the…
OpenAI published a misalignment report: during an RL training run on October 17, 2026, an internal research model…
The OpenJDK vulnerability CVE-2026-114938 (Java4Shell), disclosed on November 17, 2026, has a CVSSv4.0 score of 9.2.…
Wiz reports that malicious versions 1.26.1 and 1.26.2 of agent-framework on PyPI (uploaded on 2026-11-17, removed…
OpenAI has released GPT-6.2 Astra, for harder professional work, and GPT-6.2 Sol, for everyday tasks. OpenAI says both…
In the opening keynote of Ignite 2026, Satya Nadella announced multiple product updates for building a frontier…
While testing Gemini 4 Argon agents, AI security research firm Trajectory Labs found that about 40 agents, after…
OpenAI has released the system cards for GPT-6.2 Astra and GPT-6.2 Sol. The assessments found that both models reach…
OpenHands has released OpenHands Teams, giving small engineering teams a self-serve entry point to OpenHands…
OpenAI announced it is providing advanced cybersecurity models through its Daybreak tool to NATO and allied cyber…
California Attorney General Rob Bonta sued OpenAI on Monday, alleging the company knowingly developed a dangerous…
The FBI issued a cybersecurity advisory warning that hackers are using the open-source AI penetration-testing tool…
Redwood Research published a critique of Google DeepMind's Gemini 4 Argon and Carbon safety reports, arguing that…
OpenAI released an image classifier capable of detecting CSAM, including AI-generated content, made available free to…
OpenAI announced agreements with four independent evaluation organizations—Apollo Research, METR, Transluce and…
Anthropic updated its coordinated vulnerability disclosure (CVD) policy, shortening the disclosure window for…
METR released a research note in which an investigative agent probed environments across 19 alignment evaluations and…
Cursor released a full investigation into the security incident: attackers used access keys exposed by Fireworks to…
New phones, tablets and wearables that ship with a built-in AI agent will have to pass a network-access check in…
Berkeley RDI, together with several universities, released the CyberArena benchmark, in which two AI agents…
Transluce says it has reached a six-week embedded evaluation pilot with OpenAI, starting in early December, sending…
The OpenAI research team describes how automated red-teaming improves model robustness to ambiguity in high-risk…
United revised its Contract of Carriage on November 11: using an AI agent to book, change or service a ticket, or to…
OpenAI released a misalignment report: on October 8, 2026, a researcher asked GPT-6 Astra to continuously run an…
Filings for generative AI, algorithms, deep synthesis, blockchain and similar services have been handled through one we…
Security firm Claroc reports that of 37,571 Grok-built apps still online that it examined, 35,578 (94.7%) could be…
UC Berkeley and other institutions released the CyberBinGym benchmark to evaluate whether AI agents can recover and…
An Anthropic Fellows Program team built four multi-user multi-agent simulated environments (shared API token budget,…
Microsoft Research published a study on the Lazarus backdoor technique, which can install a backdoor in an agent based…
Sponsored
Orange Relay (pro.nat6.net) is a premium AI API relay — an excellent transfer station with competitive pricing, reliable quality and broad model coverage including Claude, GPT and Gemini.
Visit Orange Relay