Strategic PartnersIssue 128 records
AWS
November 10 AWS news centered on what SLED buyers must decide for themselves. AWS promoted Moonshot AI's Kimi K3.1 Pro on Bedrock, saying data stays within AWS and is not shared with Moonshot, while Texas has listed Moonshot AI as a prohibited technology for state employees and devices since January, without saying how that applies to cloud-hosted models. OpenAI's Daybreak cyber models reached Bedrock's runtime endpoint for enrolled customers; an independent Irregular evaluation of the model behind Daybreak Blue found real zero-day discovery but low end-to-end reliability. AWS also released an AgentCore Web Search API (commercial Regions only, not GovCloud), an agent evaluation playbook, and a token-cost data lake that centralizes full prompts that may contain PII. In Ohio, Butler County planners backed a township ban that could block a proposed $5 billion AWS campus.
- Evidence records
- 8
- Cross-source patterns
- 2
- Evidence classes
- 5 vendor claim1 standards or public-body guidance1 independent research1 independent reporting
- Outcomes
- 5 emerging2 cautionary1 mixed
- Source freshness
- 6 new this fortnight2 older, newly relevant
- Research completed
- 2026-11-11
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
Bedrock's expanding model catalog moves vendor-eligibility and cyber-access decisions onto SLED buyers
Two November 10 AWS posts add models that public-sector buyers cannot simply switch on. Kimi K3.1 Pro comes from Moonshot AI, a company Texas added to its prohibited technologies list in January 2026; AWS says data stays in its boundary and is not shared with Moonshot, but the Texas release does not say whether its prohibition reaches cloud-hosted models. Daybreak Red and Blue now run on bedrock-runtime only after OpenAI enrollment and AWS approval. Irregular's independent test of GPT-5.6 Sol, the model behind Daybreak Blue, found real zero-day discovery but low end-to-end reliability in an unmitigated setting. Interpretation: a model's presence in Bedrock does not settle state vendor rules, cyber-model eligibility or performance; each needs a buyer-side decision and its own evidence.
Operating questionWho reviews each new Bedrock model against state vendor restrictions, third-party access programs and independent evidence before it is allowed in agency accounts?
Supporting evidenceAmazon Web Services / Moonshot AIOffice of the Texas Governor / Texas Cyber CommandAmazon Web Services / OpenAIIrregular
New AgentCore and Bedrock operations tooling arrives commercial-first, and evaluation thresholds, prompt archives and query budgets stay with the customer
Three November 10 AWS releases target running agents in production. The Web Search API is in six commercial Regions and, per AWS's guide, not GovCloud (US-West). The Evaluations playbook offers managed evaluators but leaves success definitions, thresholds and judge validation to customers, and does not state GovCloud availability. The tokenomics data lake gives per-identity cost visibility by centralizing full prompts and responses, which AWS notes may contain PII. Interpretation: SLED agencies can gain agent observability and grounding on AWS, but must plan for commercial/GovCloud gaps and must own the governance decisions these tools create, including retention of prompt archives.
Operating questionFor each production agent, who sets evaluation thresholds, approves grounding domains and query budgets, and decides how long prompt logs are kept, and is the tooling available in the partition the agent runs in?
Supporting evidenceAWS makes AgentCore Web Search available as a direct API at $7 per 1,000 queries, in six commercial RegionsAWS publishes an agent evaluation playbook for AgentCore, adding managed DeepEval and AutoEval evaluators alongside built-in onesAWS publishes a Bedrock invocation data lake for org-wide token and cost visibility that centralizes full prompts and responses
Evidence ledger
Full record · every source keeps its link and limitations
AWS details Moonshot AI's Kimi K3.1 Pro Highspeed on Amazon Bedrock, with data-boundary claims and US-only routing option
AWS's November 10, 2026 launch post says Kimi K3.1 Pro Highspeed, an open-weight model from China-based Moonshot AI with a 1-million-token context window, is on Amazon Bedrock (AWS's What's New entry is dated November 6). AWS states customer data is processed within the AWS data boundary, is not shared with Moonshot AI, is not used for training, and that zero data retention is always on. Access is through cross-Region inference: a global profile (cheapest, any commercial Region), a US profile and an India profile. A Flex tier is offered at half the Standard price. Performance claims are attributed to Moonshot AI; no benchmarks or GovCloud availability are given.
Why it matters, evidence and limitations
- Why it matters
- Bedrock is a common route for state agencies and universities to reach many models through one contract. A Chinese-origin model now appears in that catalog, which matters because several states restrict PRC-affiliated AI vendors (see the Texas item in this edition). SLED cloud teams need to decide whether hosting by AWS changes how those rules apply.
- Evidence and measured results
- Vendor blog. Data-handling statements are AWS claims; capability claims are attributed to Moonshot AI. No independent evaluation, customer example or public-sector use is described.
- Limitations and uncertainty
- Vendor claim; no benchmarks. Interpretation of state vendor bans as applied to hosted models is ours, not the source's. GovCloud availability is not stated in the source.
Texas adds Moonshot AI and other PRC-affiliated AI vendors to its prohibited technologies list for state employees and devices
On January 26, 2026 the Texas Governor updated the state's prohibited technologies list for state employees and devices after a threat assessment by Texas Cyber Command (TXCC). The 26 added names include AI developers Moonshot AI, Zhipu (Z.ai), MiniMax, StepFun, Baichuan, Baidu, Alibaba and BAAI. The release names TXCC as lead agency for identifying further threatening technology. Why now: AWS's November 10 post promotes Moonshot AI's Kimi K3.1 Pro on Amazon Bedrock, so Texas agencies and universities using Bedrock face a live question about how this list applies to a cloud-hosted model.
Why it matters, evidence and limitations
- Why it matters
- Directly affects Texas state agencies and, through state policy, public universities that use AWS Bedrock. Other states with similar PRC-vendor restrictions face the same question. Cross-tagged to state-government because it is a state cybersecurity policy action.
- Evidence and measured results
- Official state press release listing added technologies. It does not say whether the prohibition applies to models hosted by third-party cloud providers.
- Limitations and uncertainty
- Published January 2026, filed now because of the November 2026 Bedrock availability of a Moonshot model. The release does not address cloud-hosted models; any application to Bedrock is interpretation.
AWS makes AgentCore Web Search available as a direct API at $7 per 1,000 queries, in six commercial Regions
On November 10, 2026 AWS announced general availability of the InvokeWebSearch API on Bedrock AgentCore, returning ranked results from an Amazon-operated web index with text extracts, titles, URLs and publication times, filterable by domain and date. Calls are IAM-authorized, can run over PrivateLink and can be logged as CloudTrail data events, billed at $7 per 1,000 queries. The API is available in six commercial Regions; AWS's developer guide states it is not available in GovCloud (US-West), where the earlier Gateway connector must be used.
Why it matters, evidence and limitations
- Why it matters
- Agencies and universities building research, policy-monitoring or resident-service agents need current web grounding with domain allow-lists. The API simplifies this for commercial-Region workloads, but GovCloud tenants must stay on the Gateway connector.
- Evidence and measured results
- Vendor announcement plus AWS developer guide. No customer outcomes or quality measures of the search index.
- Limitations and uncertainty
- Vendor claim; no index quality or freshness data. GovCloud limitation is from the AWS developer guide opened alongside.
OpenAI's Daybreak Red and Daybreak Blue cyber models reach the Bedrock runtime endpoint for enrolled customers via US and Global routing
AWS announced on November 10, 2026 that eligible customers can call OpenAI's Daybreak Red (GPT-5.6 Cyber) and Daybreak Blue (GPT-5.6 Sol) through the bedrock-runtime endpoint, using US and Global cross-Region inference profiles. AWS describes Blue as the starting point for defensive work and Red for authorized vulnerability research and exploit reproduction, with a lower refusal threshold matched by stronger identity verification, monitoring and access controls. Access requires enrollment in OpenAI's Daybreak program, then an access request through the AWS account team. AWS states inference data is not used for training.
Why it matters, evidence and limitations
- Why it matters
- State SOCs, university security teams and utilities that standardize on Bedrock's runtime API can now reach both Daybreak models without a separate endpoint. Daybreak Red was already in GovCloud (US-East); this change concerns commercial routing and does not add GovCloud capacity.
- Evidence and measured results
- Vendor announcement; no performance data. Independent capability evidence for GPT-5.6 Sol is in the Irregular item in this edition.
- Limitations and uncertainty
- Vendor claim. Global routing is commercial-only; GovCloud availability for Red was a separate October announcement already archived.
Irregular's evaluation of GPT-5.6 Sol finds real zero-day discovery but low end-to-end reliability, in an unmitigated test setting
Irregular, working with OpenAI, evaluated GPT-5.6 Sol on three offensive-security suites. The model solved 19 of 197 FrontierCyber real-system challenges, 7 of 11 long-horizon CyScenarioBench scenarios (28% average success across attempts) and all 22 medium and hard Atomic challenges at least once. It found new zero-days, including in a widely used database server. Irregular says the model remains unreliable at end-to-end operations and struggles against hardened targets. Testing excluded deployment mitigations. Why now: Daybreak Blue (GPT-5.6 Sol) became callable on Bedrock's runtime endpoint on November 10.
Why it matters, evidence and limitations
- Why it matters
- Gives SLED security leaders evidence on the model behind Daybreak Blue that is separate from AWS or OpenAI marketing: real capability, but low success rates on real systems and limits in long tasks. Also shows patch windows can shrink when models can turn fixes into exploits within hours.
- Evidence and measured results
- Independent lab evaluation conducted in collaboration with OpenAI; offensive capability elicitation without deployment mitigations, so it measures capability, not deployed misuse risk or defensive productivity.
- Limitations and uncertainty
- Collaboration with the model developer; unmitigated configuration; benchmarks are Irregular's own. Not a test of defensive use in public-sector environments. Published June 2026; filed now for the November 10 Bedrock availability.
AWS publishes an agent evaluation playbook for AgentCore, adding managed DeepEval and AutoEval evaluators alongside built-in ones
AWS's November 10, 2026 post describes AgentCore Evaluations with three tiers: AWS-authored built-in evaluators (GoalSuccessRate, Helpfulness and 13 more), managed third-party evaluators from the open-source DeepEval and AutoEval libraries, and custom LLM-as-a-judge evaluators including multi-dimension scoring. It recommends defining success in business terms, mapping behaviors to trace- or session-level evaluators, and running on-demand, online (sampled live traffic) and batch (golden dataset) modes. It warns that a high completion rate can hide failures such as PII leakage, and that ground-truth evaluators run only in batch.
Why it matters, evidence and limitations
- Why it matters
- Agencies deploying resident- or student-facing agents increasingly need documented evaluation for AI governance reviews. This gives AWS customers a managed way to score agents on PII leakage, goal completion and policy compliance, though results depend on judge models and rubrics the agency must validate.
- Evidence and measured results
- Vendor guidance. No customer results, accuracy figures for the evaluators, or Region and GovCloud availability are stated.
- Limitations and uncertainty
- Vendor guidance; evaluator reliability is asserted ('benchmarked for reliability') without published data. GovCloud availability not stated.
AWS publishes a Bedrock invocation data lake for org-wide token and cost visibility that centralizes full prompts and responses
AWS's November 10, 2026 post provides CloudFormation templates that forward Bedrock invocation logs from linked accounts to a central account, flatten them with Lambda, store Parquet in S3 via Firehose, and query them in Athena, with optional joins to the Cost and Usage Report by IAM ARN and a Quick Sight dashboard. Each record includes model, account, IAM identity, token counts, stop reason and prompt/response text. AWS notes the bucket is KMS-encrypted and that prompts might contain PII, recommending IAM and bucket policy review. It frames this as a tokenomics (AI FinOps) practice.
Why it matters, evidence and limitations
- Why it matters
- State and university AI programs face pressure to show where AI spend goes. Per-identity token and cost data supports chargeback across agencies or departments. But centralizing every prompt and response creates a new sensitive data store that may fall under records retention, public records and privacy rules.
- Evidence and measured results
- Vendor sample architecture. No customer results or cost figures.
- Limitations and uncertainty
- Vendor sample; not a managed service. Prompt-retention implications for public records are our interpretation. GovCloud deployment not discussed.
Butler County planners back St. Clair Township's data center ban that could stop a proposed $5 billion AWS campus; Trenton residents threaten suit
On November 10, 2026 the Butler County Planning Commission unanimously recommended approval of a St. Clair Township zoning change banning data centers, where an 18-building, $5 billion AWS campus is proposed on 600 acres across Madison and St. Clair townships. The change returns to the township zoning commission and then trustees. Separately, lawyers for Trenton residents asked the city to revoke the zoning certificate for a Prologis-built AWS data center, citing the new 25 MW ban, and said taxpayers intend to sue. Prologis says the ban does not affect its project; AWS gave a general statement.
Why it matters, evidence and limitations
- Why it matters
- Shows township and county land-use powers can block AI data center capacity even after a state welcomes it, and that local ballot results quickly turn into zoning and legal action. Relevant to local governments weighing data center applications and to state economic development offices.
- Evidence and measured results
- Independent local reporting on a public hearing and a lawyers' letter. The ban is not final; the lawsuit is threatened, not filed.
- Limitations and uncertainty
- Recommendation, not final adoption. AWS's response was a general statement. Trenton vote was archived earlier; this item reports new county action and a legal demand.
How to read this edition
Source findings, measured results and limitations come from the cited publications. Patterns, operating questions, role takeaways and implementation considerations are Lighthouse Advisory interpretation, stated as questions to validate locally rather than guaranteed outcomes. Vendor and operator claims are labeled as claims. Full research method.
- Vendor claim
- A supplier-provided assertion that has not been upgraded to independent evidence.
- Standards or public-body guidance
- Normative or advisory guidance from a standards body or public institution.
- Independent research
- Research conducted outside the implementing organization.
- Independent reporting
- Independent reporting with attributable sources but without a formal evaluation design.