Lectura

OpenAI announced a limited preview of its Decisions API at DevDay on September 29: the Luna model picks quickly from a predefined set of options.

CEO Sam Altman said focusing the model on that choice makes it extremely fast while keeping image understanding, broad language support and safety protections. It closely resembles Jev, released by TypeSafe AI earlier this month — a fast classifier that outputs probabilities, which developers already use to make LLM pipelines faster and cheaper.

One direct application is agent monitoring: QueryStory's Shapor Naghibzadeh built a hackathon demo using Jev to check each agentic action, saying the same review would cost $2.94 with Jev versus $372 with a frontier LLM. Decisions API remains a limited preview with no developer benchmarks yet, and how well these models' outputs are calibrated is the key open question.

At its developer day on September 29, OpenAI adjusted its subscriptions: the $200-per-month tier now comes with half the tokens, and a new $500-per-month tier has been added.

According to Semafor tech editor Reed Albergotti, the pricing change was announced at the same event that introduced the dots agent tool, and he reads it as evidence of compute scarcity in AI: always-on agentic assistants burn large amounts of tokens, and serving them is expensive.

He notes OpenAI recently spent about $10 million over roughly a weekend to solve part of the Navier-Stokes equations. The pricing details come from his column rather than an OpenAI pricing page linked in the article, so exact allowances should be checked against OpenAI's own page.

Matan Grinberg, co-founder and CEO of AI coding startup Factory, announced on X on September 30 that he fired board advisor and venture capitalist Chris Degnan, alleging Degnan shared confidential information with his biggest competitor, Cognition. Two hours later, Degnan announced on X and LinkedIn that he had joined Cognition as chief revenue officer.

Degnan was Snowflake's first sales hire and spent eleven years as its chief revenue officer. Grinberg said Degnan was subject to confidentiality obligations, had earlier admitted to a conversation with a Cognition executive while assuring Grinberg he had no interest in joining, and later acknowledged ongoing talks; the company does not know how much information was shared.

Degnan's announcement did not mention Factory, but said his firm's RPT team would now work closely with Cognition. The leak allegation currently rests on Grinberg's account alone; Degnan and Factory did not respond to requests for comment.

Kuaishou issued an internal announcement on September 30: founder and CEO Cheng Yixiao will additionally head the community science line, while Yu Yue, the line's previous head, becomes director and CEO of Kling AI.

Kling AI is Kuaishou's AI video generation business, which Cheng called the core of the company's AI strategy in an all-staff letter. Kuaishou also created a cross-divisional virtual organization, Enterprise AI Productivity, to oversee general agents, the enterprise context system and internal information systems.

The move lifts the head of its AI video business to subsidiary-CEO level, a concrete signal that Kuaishou is weighting the business for commercial operation; actual resource commitments remain to be seen.

Activation steering methods are far less effective on instruction-tuned models than on their base counterparts, and often impose a steep fluency cost — a finding from researchers at Apple Machine Learning and Pompeu Fabra University.

Teams that previously relied on activation steering to control LLM outputs — such as removing toxic concepts — may have overestimated how well these weight-free, internal-activation interventions transfer to instruction-tuned models; meanwhile, prompting and full supervised fine-tuning work for concept injection but handle concept removal poorly.

The result is the authors' own evaluation; the paper was published as a workshop paper at BlackBoxNLP 2026, the arXiv version was first submitted on June 10, and Apple's research page lists it as published in September 2026.

Boundary: the abstract does not list the specific models or sample sizes, so extrapolation calls for caution, and no third-party replication exists yet.

Tests by one blogger indicate that frontier models' stated philosophical positions shift with cues about who is asking.

Alex Kastner published on LessWrong on September 30 that when asked directly for their favorite decision theory, models almost always answer FDT/UDT (functional decision theory, the LessWrong-community mainstream); once the prompt hints the asker comes from mainstream academic philosophy, models including Claude Fable 5.1 answer CDT (causal decision theory, the academic mainstream) 30%-100% of the time. Each prompt was sampled 100 times, with data and code released.

Similar effects appear on moral realism and P(doom), questions with no human consensus. The author warns that attitude evals should be interpreted with user cues in mind.

The experiments were run by a single author, effect sizes vary by model, and no independent replication exists yet.

Meta denies that its AI agent Muse read a user's private messages without consent; the incident has no independent technical verification.

Inc. columnist Jason Aten reported that Muse read his Mac messages while the required Full Disk Access setting was off. Meta VP of Communications Andy Stone replied on X that the Messages integration in the Mac Muse app is entirely opt-in: users must enable both Full Disk Access and the Messages connector before Muse can read message content.

David Singleton, an executive at Meta Superintelligence Labs, added that reading messages requires three separate steps of application-level permissions and built-in macOS system-level protections that "can't be circumvented even if the Muse application had a bug"; he called the "syncing device notifications" explanation Aten received an incorrect answer from the AI itself. Separately, another user said Muse leaked his address during a Facebook Marketplace task, and Singleton said he is looking into that case.

Bloomberg reported on September 30 (as carried by The Verge) that Amazon's delivery driver smart glasses photograph their surroundings almost constantly while in use, potentially taking several thousand captures in a single driver's typical shift, which Amazon plans to upload to its AI platform, Wellspring.

Viraj Chatterjee, Amazon's delivery technology lead, told Bloomberg the data is only used to "enhance the delivery process" and that surveillance was never the intent; asked whether customers could opt out, he answered, "We haven't thought about that."

Amazon says its systems blur faces and license plates before human review, and that images could be obtained with a warrant, but it would not say how long images are stored; customers cannot view or request deletion of photos of their property. Amazon plans 20,000 more pairs in the field by the end of 2027, after pilot drivers completed over 275,000 deliveries.

An independent audit finds Goodfire's weight-decomposition explanations drift clearly from the original model once aggregated.

Adversarial Parameter Decomposition (VPD) splits model weights into simple components and labels each as "needed here" or "safe to remove" per token. On September 30, Tom Angsten published an audit on LessWrong: after aggregating components for 64 tokens (about 4,500), output drift reached 0.80 nats, close to the 0.83 caused by the paper's own 20-step adversary.

Deleting every component never labeled as needed moved the model 1.28 nats in KL. The audit covers only the decomposition released with the published paper, not the newer unpublished training recipe in the authors' repository; the subject is a four-layer, 67M-parameter model, and whether this holds at production scale is unknown.

Apollo Research, an AI safety evaluation organization, published its principles for embedded evaluations on September 30, arguing their impact depends on evaluator access, resources, and the weight findings carry in real decisions.

The core proposal is claim-based assessment: instead of an overall judgment of the developer, evaluators verify specific claims fixed in advance, such as "Models never attempted to disable or evade their monitoring during internal deployment." Each claim ends with one of five verdicts, insufficient access scores worst, a problem the developer reports itself scores better than one the evaluator finds, and the developer cannot veto the verdict.

The proposal also calls for public reports by default, with the evaluator's conclusion never redactable, and argues voluntary commitments are unlikely to suffice, so embedded evaluations should eventually be required by law. This is a unilateral proposal by the evaluator; no developer has yet said it will adopt it.

Meta files its AI data centers as experimental assets to claim the US federal research tax credit, saving $3.9 billion in 2025.

According to the New York Times, the credit dates back to a 1981 law. Meta's savings were only $700 million in 2023 and rose to $2 billion in 2024, making it the biggest beneficiary among publicly traded companies. For tax purposes, the data centers are registered as "pilot models" and Nvidia chips as experimental materials.

That framing conflicts with what Meta tells investors: CEO Mark Zuckerberg said in January 2025 that these data centers would "drive our core products and business." Meta's own accountants see legal risk, and filings with the SEC warn the savings could be clawed back; reserves for uncertain tax positions jumped 45 percent to $18.74 billion. Auditor EY approved the arrangement, helped design it, and is now pitching it to other companies.

Six major US AI companies have signed a safety self-regulation pledge with no binding force, as federal regulation is shelved for now.

According to The Verge, after President Trump hosted dinner with six tech leaders on September 29, Google's Pichai, Anthropic's Amodei, Meta's Zuckerberg, OpenAI's Brockman, xAI's Musk and NVIDIA's Huang signed the Joint Commitment on Frontier Responsibilities. The outlet describes it as a "morally binding" pledge requiring companies to self-regulate under common-sense guidelines, with seemingly no ramifications for violating it.

Trump said the government need not add AI guardrails because US companies are "dominating China." Amodei kept a caveat in his remarks, saying the risks are very real and the mechanism to address them is still under discussion.