AI Safety & Alignment

Today's Stories

1. 1

   Independent researcher shares recurrent AI architecture DABSN, seeks collaborators

   A researcher working independently has released DABSN (Dynamic Adaptive Bias State Network), a recurrent architecture with public preprints, PyTorch, C++, and Triton code. The researcher has also trained a 24M-parameter language model on 1B pretraining tokens using a GPT-2 tokenizer and is now writing a second paper focused on language modeling and long-context behavior. Open-source recurrent architectures with public implementations allow the broader research community to reproduce, verify, and build on novel approaches outside large corporate labs. The researcher's unexpected results on language modeling suggest the architecture may offer alternative efficiency or capability profiles worth investigating.

   The researcher is actively seeking collaborators for independent reproduction, evaluation, and the next language-modeling paper—a common early stage for open research when validation and scaling require broader participation.

2. 2

   NeurIPS 2026 launches inaugural workshop on real-time conversational AI

   The RTCA (Real-Time Conversational Agents) Workshop is being launched as the inaugural workshop at NeurIPS 2026, held in Sydney, Australia on 11 or 12 December 2026, with a call for papers and demos now open. Conversational AI has moved beyond text chat into voice, video, and embodied avatars that must operate in real time—streaming while listening and watching simultaneously. This requires handling latency, turn-taking, interruptions, and cross-modal alignment, which are fundamentally harder problems than offline text generation and represent a shift in how AI systems must interact naturally with users.

   The workshop focuses on streaming speech, video, and language generation; naturalness in real-time interaction; and evaluation of live systems. Submissions and demos are being accepted through the workshop website at https://rtcaneurips26.github.io/.

3. 3

   Half of enterprises ship AI agents that fail in production; most skip human review

   A survey of 157 enterprises found that half have already deployed an AI agent that passed their internal evaluations but then failed when used by customers in production. Only one in twenty fully trusts automated evaluation today, and the most-cited weakness is that evaluations do not align with real-world outcomes. Organizations are granting AI agents more autonomy while losing confidence in the evaluations meant to control that autonomy. Two-thirds of surveyed enterprises already allow, or are actively engineering toward, deploying agent changes to production on automated evaluation alone — with no human in the loop — creating what the research calls an "evaluation gap" between the autonomy granted and the trust placed in safeguards.

   The gap between enterprise confidence in automated evaluation and their willingness to deploy without human oversight suggests a structural mismatch: companies are shipping faster than their testing can validate, and many are discovering failures only after customers encounter them.

4. 4

   Cohere VP: Enterprise AI sovereignty needs full control of agent stack

   Rachad Alao, vice president of product engineering at Canadian AI startup Cohere, spoke at VB Transform 2026 in Menlo Park about building AI agent systems while keeping sensitive data and infrastructure under enterprise control. Alao, who previously led responsible AI teams at Google and Meta, argued that AI sovereignty requires more than just running an open model behind a corporate firewall. Banks, hospitals, and governments operating mission-critical systems need tight control over where their data resides and the ability to switch vendors without being locked into a single AI provider. For enterprises handling sensitive information, partial measures—downloading a model or using a firewall—are insufficient; true sovereignty demands control over the entire AI agent stack.

   The talk reflects growing concern among large organizations that relying on third-party AI infrastructure and vendors could expose proprietary operations or create dependency. Cohere's emphasis on full-stack control suggests the startup is positioning itself to serve enterprises unwilling to cede operational autonomy to larger cloud or AI vendors.

5. 5

   OpenAI launches GPT-Red, automated safety system using self-play

   OpenAI has introduced GPT-Red, an automated red teaming system that uses self-play to improve AI safety, alignment, and robustness against prompt injection attacks. Red teaming—deliberately probing AI systems for weaknesses—is a core part of AI safety work. Automating this process with self-play (where the system tests itself) may allow OpenAI to identify and fix vulnerabilities faster and more systematically than manual testing alone, strengthening the safety of deployed models.

   The system focuses on three key areas: AI safety, alignment, and prompt injection robustness. How effectively GPT-Red scales to catch edge-case failures in production models will be a measure of its real-world impact.

6. 6

   OpenAI proposes 'reverse federalism' for US AI governance

   OpenAI has outlined a governance approach it calls 'reverse federalism,' in which state-level laws can help construct a national framework for AI safety and democratic principles. Currently, AI governance lacks a unified national standard in the US, and OpenAI's proposal suggests state experimentation could inform federal policy rather than federal mandates flowing down—potentially offering a faster, more adaptable path to establishing safety norms across the industry.

   This framing mirrors debates in other sectors where state innovation precedes federal action; whether Congress and state legislatures adopt this model will shape how AI safety standards emerge over the coming years.