GenAI Daily - October 10, 2026: Microsoft's Decision-1 Model, OpenAI Revenue Correction, Anthropic's Free OSS Scanner
Top Stories
Microsoft Ships Decision-1, a Small Model That Scores Fixed Answer Options
Microsoft CEO Satya Nadella unveiled Microsoft-Decision-1 on Friday.
Microsoft says it is built for routing, classification, prioritization, verification and workflow control.
Given a fixed set of answer options, it returns a calibrated probability score for each one. It supports yes/no, multiple-choice and rating formats, plus rubric-based grading of AI responses and agent actions, through a structured API call.
Microsoft post-trained Qwen3.5-9B for single-pass scoring and plans to rebase the model on Microsoft AI (MAI) and OpenAI models.
Pricing is $0.042 per million input tokens, with output tokens free.
Some outlets reported $0.42, so check the Foundry listing before budgeting.
The Foundry catalog lists it as generally available for text classification, with JSON output and a roughly 32.8k context window.
OpenRouter access is "coming soon," with no date given.
Microsoft says it ran 35 times faster than GPT-6 Sol and 4.5 times faster than Quyet-1.0-Large in internal testing.
Those figures are Microsoft's own and have not been independently verified.
Xbox research used it to classify more than 10,000 pieces of gamer feedback, with quality similar to GPT-6 Sol, Microsoft says.
Microsoft Decision-1 pricing and details (FourWeekMBA) | Foundry listing details (WindowsForum)
Why it matters: Microsoft now sells a decision model at the same posted input price as TypeSafe's Jev, so teams running LLM calls for routing and classification have a cheaper option to benchmark against their own data.
OpenAI's Annualized Revenue Is About $50B, Not $70B, and Markets Reacted
The Financial Times reported that OpenAI's annualized revenue is about $50 billion as of September, roughly $20 billion below figures previously reported to investors.
A source familiar with the documents said the $70 billion number did not come from OpenAI. It likely came from firms trying to compare OpenAI with Anthropic, whose figures include gross revenue from cloud providers.
The FT's reading of investor materials puts OpenAI's July figure near $30 billion, with growth of more than 70% since.
At $50 billion, OpenAI would sit below Anthropic's $65 billion in annualized sales.
The Nasdaq Composite slid on Thursday as the AI trade came under pressure.
Oracle fell more than 5%, and Nvidia and AMD lost nearly 3% and almost 4%.
Analysts frame this as a methodology correction rather than a demand drop.
That framing did not stop AI-related stocks from pulling back.
CNN | Yahoo Finance (TechCrunch coverage)
Why it matters: Vendors report "annualized revenue" on different bases (gross vs. net of cloud partners), so treat run-rate claims as marketing inputs when assessing a provider's staying power.
Anthropic Opens OSS Scanner, a Free Claude Mythos Vulnerability Service for Open-Source Maintainers
Anthropic unveiled OSS Scanner as an opt-in vulnerability scanner for the open-source ecosystem.
Participating projects get periodic security scans by Anthropic's strongest models at no cost.
Outputs are fully model-generated, with no human review or triage.
Reports may include an explanation, a reproducer and a proposed patch, and maintainers must validate and triage the findings.
In testing, penetration testers reviewed 97 critical- or high-severity findings across 48 projects. 85 of them (88%) met Anthropic's coordinated-disclosure standard.
The models identified more than 29,000 candidate vulnerabilities in the six months before launch.
Core maintainers apply through the GitHub repo with a pull request that adds a project.yaml containing the repo address and a contact email.
Projects also need a Dockerfile, and everything for building and testing must download during setup because the audit cannot reach the internet.
Anthropic separately offers maintainers free Claude Max 20x subscriptions through Claude for OSS to help remediate issues.
The Hacker News | Anthropic Frontier Red Team
Why it matters: Teams that depend on or maintain open-source components should expect more AI-generated, untriaged vulnerability reports, so plan triage capacity and patch workflows now.
Key Developments
UPDATE: Fired OpenAI Safety Researchers Publish Open Letter Disputing Misconduct Claims
Jasmine Wang, Tomek Korbak and Mikita Balesni, the three researchers OpenAI fired last week, published an open letter denying that they mishandled sensitive information. They warned that the dismissals create a chilling effect on safety culture.
They denied involvement in a leak to The Information about less monitorable model architectures, and denied engaging with external parties outside their job mandates.
OpenAI responded that the decisions "were not about raising safety concerns or speaking out."
It also reaffirmed its commitment to independent safety audits.
Yahoo News (TechCrunch) | Yahoo News
Impact: Buyers with vendor-governance or AI-risk review requirements should note the dispute over how OpenAI handles outside safety collaboration and audits.
Salesforce Winter '27 Turns Agentforce On by Default Across Eligible Orgs
Production upgrades roll out in waves, with main release weekends on October 3 and October 10, 2026.
Eligible editions receive Agentforce enablement automatically at no additional cost, and Salesforce has indicated the setup toggle will be removed later in the release cycle. Enablement makes the tools available but does not activate any agent or create consumption charges by itself.
AI Skills in Agentforce Coworker and Agent Optimizer are both slated for general availability in October 2026.
Impact: Salesforce admins should review permissions, data access and usage-billing settings before agents are switched on in their orgs.
ASOS Breach Claim Points to Simon AI, a Snowflake Cortex-Based Marketing Tool
The group behind the ASOS hack told the BBC it breached Simon AI, a service integrated with Snowflake, and exfiltrated customer data. They said it did not include payment card information.
Snowflake said the issue did not result from any vulnerability or misconfiguration in its platform and that no remediation is required for Snowflake customers.
Simon AI is an agentic marketing platform built on Snowflake Cortex AI, and Monetate acquired it in July.
ASOS said the attackers impersonated a trusted contact to steal login credentials and reach third-party customer communication platforms.
The link between Simon AI and the stolen data remains the attackers' claim.
Infosecurity Magazine | BankInfoSecurity
Impact: The exposure sits in credentials and third-party agent tooling around the data warehouse, so audit which AI vendors hold standing access to your data platform.
Anthropic Rewrites Usage Policy, Effective November 12
Anthropic's updated policy prohibits "sustained and needless abusive or cruel behavior" toward its models, and takes effect November 12.
The rewrite also tightens rules on deception, weapons and surveillance.
The Register notes it also covers election interference.
Impact: Compliance teams building on Claude should compare the new terms with their products' acceptable-use language before the effective date.
Product Launches
Claude Managed Agents Dynamic Workflows (Beta)
For work with many pieces, such as reviewing hundreds of documents, a Managed Agents agent can now write a workflow, which is a program that runs many agents in phases and combines their results.
Enable it with the multiagent field set to type multiagent_20261001 and workflows enabled.
Runs are tracked through workflow_run events on the session's event stream.
Teams doing bulk document review can test fan-out without building their own orchestrator.
Why it matters: Teams doing bulk document review or other work with many parts can now test fan-out across many agents without building their own orchestrator, which lowers the effort needed to run large multi-step jobs.
Funding & Deals
UPDATE: TypeSafe AI Raises $870M Series A
San Francisco-based TypeSafe AI raised an $870 million Series A at a $7.5 billion valuation, with Sequoia Capital, existing investor DCVC and angels participating. a16z general partner Martin Casado joins the board.
The company makes Jev, a model that takes typed questions, evaluates them against a supplied state and returns structured results without generating text.
TypeSafe says about a third of the Fortune 500 use Jev but has not named customers.
It came out of stealth on September 15 with a $40 million seed.
The close follows earlier reports of investor interest in a larger round. Led by Andreessen Horowitz.
Why it matters: A large Series A for a model that returns structured results instead of generated text signals investor appetite for narrow, decision-focused AI, and it puts TypeSafe in direct competition with Microsoft's new Decision-1.
Broadcom Arranges $50B+ Debt Financing for OpenAI Custom Chips
Broadcom is in early talks on more than $50 billion in debt financing to help OpenAI fund purchases of the custom AI chips the two companies are jointly developing.
No deal is expected before the end of the year.
It would support a program agreed in October 2025 that calls for 10 gigawatts of accelerators by late 2029.
The shift to bespoke private financing suggests AI buildout costs are outrunning what traditional debt markets can absorb.
Apollo and Blackstone are among the lenders approached.
Seeking Alpha | CryptoBriefing
Why it matters: Financing on this scale shows how costly custom AI chip buildouts have become, and it could shape how quickly OpenAI can secure the hardware capacity it plans for through 2029.
Tomorrow's Watch List
- Microsoft-Decision-1 on OpenRouter, which has no date yet, and independent benchmarks to check the self-reported speed claims.
- Reflection AI's Beam: the company says weights and full technical details arrive this month.
- Salesforce AI Skills, Agent Optimizer and Slackforce general availability, all targeted for October.
- OpenAI's response to the researchers' open letter, plus any further investor disclosures on revenue.
*Related reading: Check out this week's [Deep Insights analysis] for strategic context on these developments.
