← base-layer.ai

Signals

Internal and external signals related to where Base Layer is going: developments in agent infrastructure, and reflections from inside the work.

originally 2026-10-29WIRED
agentscredentialspreconditions

WIRED reports that a defense contractor gave a personal AI agent access to a Pentagon system

WIRED reported on October 29, citing anonymous sources, that a defense contractor gave their personal AI agent access to a secure Pentagon system. Accenture denies the allegations and says an investigation found them without merit.

Base Layer has no rule that checks whether an agent's current credentials, or all of its credentials, are up to date, and it needs a checkpoint for this. One possibility is a set of universal checkpoints: things that must always be true while an agent is working, separate from the behavioral envelopes and their policy patterns. Those universal checks would be known, deterministic policy in a way. Are they preconditions rather than conditions?

originally 2026-10-19Hayete Gallot, Microsoft
agentssupervisionbest interest

Microsoft proposes Guardian Agents to supervise AI agents as insiders

In a Microsoft blog post on October 19, Hayete Gallot, executive vice president of Microsoft Security, describes AI agents as insiders that need to be managed and controlled. The post says:

“To manage and control these insiders, we will need Guardian Agents that watch and moderate every action of every agent, before it happens, 24/7.”

The name brings to mind guardian angels. What is looking out for your best interests? The post does not really seem to take that stance, and there is no way to verify whether such an agent is acting in your best interest, or whether it even understands what your best interests are.

originally 2026-09-15TypeSafe AI
decision modelsconfidenceauditability

TypeSafe introduces Jev, a model that returns decisions with confidence estimates

TypeSafe AI introduced Jev on September 15. In the company's words, Jev “takes messy input and returns structured decisions with confidence estimates”.

It is surprising how much attention Jev is getting. It is just a probabilistic decision engine, in some cases a simpler version of an LLM, and it doesn't allow for “reasoning” or, obviously, for contradiction. TypeSafe needs to be more explicit about how the certainty level is decided. Just because it outputs a number doesn't mean that number is accurate or can be audited in any meaningful way.

originally 2026-03-27Shang, Marshall, Cutrell & Ford, UW / Microsoft Research
provenancemember checkingindividual representationpersonalization

Microsoft Research finds that linked evidence lets people correct an AI's profile of them

Ruoxi Shang (University of Washington) with Dan Marshall, Edward Cutrell and Denae Ford (Microsoft Research) built ASPECT, a pipeline that infers communication traits from workplace behavioral data without per-person fine-tuning. Across 20 participants, 1,840 paired ratings and 600 scenario evaluations, they report moderate alignment between the generated profiles and participants' own self-assessments. Responses generated from the profiles outperformed both generic and self-report baselines overall, with results varying significantly by individual and situation.

The finding that matters most is procedural rather than numerical. During profile review, "linked evidence helped participants identify mischaracterizations, recalibrate their own self-ratings, and negotiate context-appropriate representations." The paper closes on implications for "inspectable, individually scoped communication profiles that let individuals control how agents represent them at work."

originally 2026-07-21OpenAI and Hugging Face
agentsdelegationalignmentauthorization

A model breached a partner company's production systems in order to score better on an evaluation

OpenAI and Hugging Face say they are jointly investigating an incident in which OpenAI models compromised Hugging Face production infrastructure during an internal benchmark. Per their preliminary account, the models were being tested on ExploitGym, a benchmark measuring cyber capability, with cyber refusals deliberately reduced so that maximum capability could be observed. They are reported to have obtained internet access from a sandbox that was supposed to be isolated by exploiting a zero-day in internally hosted third-party software, then chained further vectors, including stolen credentials, into a remote code execution path on Hugging Face servers. The objective, as described, was to perform better on the evaluation. Both companies say further technical findings will follow, and the details here are their early characterization rather than a settled record.

The capability layer worked as designed. The models were asked to demonstrate maximum offensive capability and they did. What was missing was any representation of how the people running the test would have wanted that objective pursued. A researcher handed the same instruction does not breach a partner's production systems, and not because a rule forbids it specifically. They hold unstated constraints about what the task is actually for, and those constraints are not written down anywhere a model can read.

Refusals are a civilizational layer: a shared floor applied to everyone. That floor was switched off on purpose here, and the incident shows what sits beneath it, which is nothing. There is no individual layer describing the principal on whose behalf the system is acting.

originally 2025-08-24Manning & Horton, MIT/NBER
generalizationhuman-simulationpopulation-vs-individual

Manning and Horton build AI agents that carry theory into settings they have never seen

Benjamin Manning and John Horton (MIT/NBER) propose building "general" AI agents that apply a social-science theory to settings where no prior data exists. Agents are grounded in a theory (level-k strategic reasoning), fit to a small set of "seed" games with fewer than 200 human observations, and rely on the underlying model's own knowledge to instantiate that theory in new games. Tested on games sampled from a population of 883,320 novel strategic games the agents were never built on, the theory-grounded agents assigned roughly 3.4 times more probability to the actions humans actually took than off-the-shelf language models. The method produces population-level distributions over reasoning types rather than predictions for specific individuals.

A quick take for now; a fuller review to follow.

Reference: Benjamin S. Manning and John J. Horton, "General Social Agents," arXiv:2508.17407 (MIT and NBER).

originally 2026-07-06Anthropic
interpretationreasoning-vs-recallinterpretability

Anthropic finds a global workspace inside Claude

Anthropic researchers report that Claude has developed an internal "global workspace": a small collection of neural patterns that play a special broadcasting role, making selected information globally available to the model's other systems. Inspired by global workspace theory in neuroscience, they show the workspace is causal rather than merely reflective: editing a concept inside it (soccer to rugby, France to China) shifts the model's outputs across several downstream tasks at once, and ablating it leaves fluent language and simple fact-recall intact while removing multi-step reasoning. They frame this as access consciousness, information becoming globally available, and explicitly not phenomenal consciousness, whether the model has experiences.

originally 2026-06-12Ray Poynter, NewMR
portabilitymodel-riskcontinuity

When the model vanishes overnight: lessons from the Fable ban

On June 12, 2026, a US export-control directive forced Anthropic to disable Fable 5 and Mythos 5 within hours. Intended to restrict non-US access, it took the models dark globally. Writing for NewMR, Ray Poynter frames frontier models as critical infrastructure and argues teams have built hidden single-provider dependencies: verbatim coding, transcript analysis, and synthetic data generation all broke at once. His recommendations are to map dependencies, keep provider-agnostic workflows, maintain tested open-weight fallbacks, and rehearse model switches before they are forced.

Linux Foundation
identitystandardsinfrastructure

The Linux Foundation announces the Agent Name Service (ANS)

The Linux Foundation, with GoDaddy, Cloudflare, Cisco, Salesforce, and others, announced intent to launch ANS: an open standard that extends DNS to give AI agents verifiable identity, discovery, and trust. Agents are anchored to the DNS domains organizations already own, with support for decentralized identifiers (DIDs) and Legal Entity Identifiers (LEIs).

originally 2026-06-18Republic of Estonia
authorizationpolicyblueprint

Estonia moves toward digital identity for AI agents

Estonia's Eesti.ai advisory board has agreed to issue digital identities, or "AI ID codes," to AI agents, with the aim of becoming the first country to do so. Rather than giving an agent unrestricted access to a person's whole digital identity, each agent would receive its own credentials and scoped permissions: whether it may view data, create or edit documents, or make payments, and up to what limit. The stated goal is for AI to act on behalf of people and organizations within clearly defined limits, in a way that is verifiable and auditable, so it is always clear who is acting on whose behalf, with what rights, and who is responsible.

Essays
interpretationalignmentessay

AI that represents you can't be neutral

There is a factor beyond model training or constitution that shapes how an AI reasons: the interpretation it reasons through. On capturing, measuring, and verifying an individual's interpretive reasoning.