# forck.live — 5–11 October 2026

> Live feed of AI lab newsrooms. First-party announcements, traced to the source.

forck.live follows what companies announce about their own products. Press may
corroborate an official story; it never leads and never appears as an item of its
own. A missing date stays missing. A title is the publisher's words — where a
company does not publish in English, an English rendering is listed separately
as `Title in English` and named as this site's translation, never as theirs.

This file is the week in progress: every confirmed first-party
announcement of ISO week 2026-W41, 5–11 October 2026, newest first. The
window never moves, so this URL always names the same seven days — but they
have not all happened yet, so the file is still filling: announcements are
appended as they land, and it will answer with more of them later today. Fetch
it again rather than reading a cached copy as the finished week.

Every entry is one first-party announcement, and carries the same eight fields
in this order before any others:

    Company · Announced · Category · Coverage · Announcement · Group · Source · Record

`Source` is the company's own page the announcement was traced to and `Record`
is this site's page for it. Both are safe to cite.

`Coverage` is a number of outlets, or `not counted` — which means this record
reached no count of its own, never that nobody wrote about it.

`Category` is this tracker's own analysis, and is authoritative wherever it is
stated. `not stated` means either that nothing has analysed the post yet or
that the analysis placed it outside this taxonomy; neither is a verdict about
the announcement.

`Announcement` is whether the post is news of something the company did, rather
than a how-to, a customer story, a changelog listing or a staffing note.

`Group` is which of this record's groups the entry falls in, carried on the
entry itself so it holds where the groups are headings and where they are not.
One of: `models`, `covered`, `announcements`, `routine`.

Anything after `Record` is held detail that only some announcements carry.

- This week's page: https://forck.live/week/2026-W41
- The current issue, and the catalogue of the other weeks: https://forck.live/briefing.md
- Catalogue: https://forck.live/llms.txt

## Announcements

### Hugging Face — Falcon-Emirati: When an LLM Learns the Dialect, the Culture, and the Nuance

- Company: Hugging Face (huggingface.co)
- Announced: 2026-10-06T06:44:39+00:00
- Category: new-model
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://huggingface.co/blog/tiiuae/falcon-emirati
- Record: https://forck.live/items/16162-falcon-emirati-when-an-llm-learns-the-dialect-the-culture-and-the-nuance
- Subject: Platform

TII released Falcon-Emirati-7B, a dialect-specialized model built on Falcon-H1-Arabic that understands and generates Emirati Arabic with cultural context. The model uses the Falcon-H1 hybrid architecture combining Mamba and Transformer attention, and was trained on authentic Emirati web data, MSA cultural content, and synthetic data guided by glossaries. The 7B parameter variant was chosen as the best balance of quality and cost for dialect adaptation.

### Amazon — Introducing GLM 5.3 on Amazon Bedrock

- Company: Amazon (amazon.com)
- Announced: 2026-10-05T23:25:17+00:00
- Category: model-update
- Coverage: not counted
- Announcement: yes
- Group: models
- Source: https://aws.amazon.com/blogs/machine-learning/introducing-glm-5-3-on-amazon-bedrock/
- Record: https://forck.live/items/16107-introducing-glm-5-3-on-amazon-bedrock
- Subject: Bedrock / Nova

Amazon announced that GLM 5.3, a 753B-parameter mixture-of-experts model from Z.ai (Zhipu AI), is now available on Amazon Bedrock. The model is optimized for coding and long-horizon agentic tasks, and Z.ai reports it shows notable cyber security capabilities. On Bedrock, it is accessible through fully managed APIs with cross-Region inference, prompt caching, and service tiers for eligible enterprise customers.

### Perplexity — Accurate sandbox extraction usage and costs

- Company: Perplexity (perplexity.ai)
- Announced: 2026-10-05T21:42:37+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://docs.perplexity.ai/docs/resources/changelog#october-2026
- Record: https://forck.live/items/16101-accurate-sandbox-extraction-usage-and-costs
- Subject: Sonar

Perplexity's Agent API now reports token usage and cost for sandbox extraction calls, broken down per model and including cache and reasoning-token dimensions. The update corrects response-level reporting without changing pricing or backfilling historical data.

### Perplexity — Image Search

- Company: Perplexity (perplexity.ai)
- Announced: 2026-10-05T21:42:37+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://docs.perplexity.ai/docs/resources/changelog#october-2026
- Record: https://forck.live/items/16100-image-search
- Subject: Sonar

Perplexity introduced an image_search tool that searches the web for images and returns typed results with image and source-page URLs. Users can apply result limits and domain, format, and safe-search filters. Successful calls cost $2.50 per 1,000 invocations, while failed calls incur no tool cost.

### Perplexity — Native document input

- Company: Perplexity (perplexity.ai)
- Announced: 2026-10-05T21:42:37+00:00
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://docs.perplexity.ai/docs/resources/changelog#october-2026
- Record: https://forck.live/items/16099-native-document-input
- Subject: Sonar

The Agent API now accepts PDF, DOC, DOCX, TXT, and RTF documents as input_file content. Send a document inline with file_data or provide a public HTTPS file_url . For inline documents, continue the conversation with previous_response_id without uploading the file again. See Working with Files .

### Perplexity — New: Decisions API

- Company: Perplexity (perplexity.ai)
- Announced: 2026-10-05T21:42:37+00:00
- Category: api-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://docs.perplexity.ai/docs/resources/changelog#october-2026
- Record: https://forck.live/items/16098-new-decisions-api
- Subject: Sonar

Perplexity released the Decisions API, which answers questions about text, JSON, or images with probabilities instead of generated text, using the pplx-decider-v1-27b model for yes-or-no, choice, or scored answers. Input costs $0.04 per million tokens, with free output tokens and no per-request fee.

### Google Research — Open and Emergent Problems in Agentic Privacy and Security: A Contextual Angle

- Company: Google Research (research.google)
- Announced: 2026-10-05T21:08:31+00:00
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://research.google/blog/open-and-emergent-problems-in-agentic-privacy-and-security-a-contextual-angle/
- Record: https://forck.live/items/16096-open-and-emergent-problems-in-agentic-privacy-and-security-a-contextual-angle
- Subject: Research

Google Research published a workshop report identifying open research problems in agentic privacy and security. The report applies Contextual Integrity theory to define privacy as appropriate information flow and proposes a multi-layered approach including a contextual policy engine to evaluate the appropriateness of agent actions before execution.

### Amazon — Supercharge regulated workloads with Claude Code and Amazon Bedrock

- Company: Amazon (amazon.com)
- Announced: 2026-10-05T17:25:20+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://aws.amazon.com/blogs/machine-learning/supercharge-regulated-workloads-with-claude-code-and-amazon-bedrock/
- Record: https://forck.live/items/16074-supercharge-regulated-workloads-with-claude-code-and-amazon-bedrock
- Subject: Bedrock / Nova

Amazon published a guide on using Anthropic's Claude Code agentic coding tool with Amazon Bedrock in AWS GovCloud (US) Regions. The post covers setting up Claude Code with Claude Opus 5.5, Claude Sonnet 5.5, and Claude Sonnet 5 models on Bedrock, which hold FedRAMP Class D certification and DoD IL4/IL5 authorization. It provides step-by-step instructions for configuring the environment, including prerequisites and installation commands.

### Amazon — New agent skill: Amazon SageMaker optimized generative AI inference for your coding agent

- Company: Amazon (amazon.com)
- Announced: 2026-10-05T17:23:19+00:00
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://aws.amazon.com/blogs/machine-learning/new-agent-skill-amazon-sagemaker-optimized-generative-ai-inference-for-your-coding-agent/
- Record: https://forck.live/items/16075-new-agent-skill-amazon-sagemaker-optimized-generative-ai-inference-for-your
- Subject: Bedrock / Nova

Amazon SageMaker AI introduced the aws-ai-ml skill, available through the Agent Toolkit for AWS, which gives coding agents such as Kiro, Claude Code, and Codex expertise in inference optimization and benchmarking. The skill allows agents to benchmark endpoints, recommend deployment configurations, compare performance runs, and generate executable SageMaker Python SDK v3 code. It can be installed locally or used within Amazon SageMaker Studio.

### Perplexity — Low, medium, and high presets use Fast Search

- Company: Perplexity (perplexity.ai)
- Announced: 2026-10-05T17:15:02+00:00
- Category: pricing-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://docs.perplexity.ai/docs/resources/changelog#october-2026
- Record: https://forck.live/items/16073-low-medium-and-high-presets-use-fast-search
- Subject: Sonar

Perplexity changed the low, medium, and high presets of the Agent API to use Fast Search instead of standard web search, reducing the price of web_search calls under those presets from $2.50 to $1.00 per 1,000 invocations and making search about 800 ms faster. The xhigh preset remains unchanged.

### Krea — introducing Krea in After Effects.

- Company: Krea (krea.ai)
- Announced: 2026-10-05T16:00:02+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://x.com/krea_ai/status/2107138756507763130
- Record: https://forck.live/items/16059-introducing-krea-in-after-effects
- Subject: Announcements

Krea announced that its Krea Agent can now be used inside Adobe After Effects to create motion graphics, resize videos, and handle sound and images.

### Amazon — Making Amazon Quick enterprise-ready: Automated, auditable cross-account resource promotion

- Company: Amazon (amazon.com)
- Announced: 2026-10-05T15:56:28+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://aws.amazon.com/blogs/machine-learning/making-amazon-quick-enterprise-ready-automated-auditable-cross-account-resource-promotion/
- Record: https://forck.live/items/16058-making-amazon-quick-enterprise-ready-automated-auditable-cross-account
- Subject: Bedrock / Nova

Amazon published a guide for automating cross-account promotion of Amazon Quick resources (chat agents, action connectors, knowledge bases, flows, and spaces) using a sample MCP server hosted on Amazon Bedrock AgentCore. The solution is idempotent, auditable, and copies permissions faithfully by reading source resources via the Amazon Quick API and recreating them in a target AWS account. The full source code is available in the aws-samples repository.

### Amazon — Agentic retrieval with LangChain and Amazon Bedrock Knowledge Bases

- Company: Amazon (amazon.com)
- Announced: 2026-10-05T15:53:56+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://aws.amazon.com/blogs/machine-learning/agentic-retrieval-with-langchain-and-amazon-bedrock-knowledge-bases/
- Record: https://forck.live/items/16054-agentic-retrieval-with-langchain-and-amazon-bedrock-knowledge-bases
- Subject: Bedrock / Nova

Amazon Bedrock Managed Knowledge Base now supports agentic retrieval, which breaks a multi-part question into sub-queries, runs them, judges whether it has enough evidence, and searches again if it doesn't. The langchain-aws package exposes both agentic and standard retrieval, allowing LangChain applications to use either method. The AgenticRetrieveStream API runs a planning loop and streams trace events back to the user.

### Amazon — Downgrading user roles in Amazon Quick

- Company: Amazon (amazon.com)
- Announced: 2026-10-05T15:51:38+00:00
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://aws.amazon.com/blogs/machine-learning/downgrading-user-roles-in-amazon-quick/
- Record: https://forck.live/items/16055-downgrading-user-roles-in-amazon-quick
- Subject: Bedrock / Nova

Amazon published a guide on downgrading user roles in Amazon Quick, covering why to downgrade (least privilege and cost savings), prerequisites, and two methods: manual deletion-and-recreation and using the AWS CLI. The console does not support direct downgrades from Admin or Author to Reader, so the CLI step-down sequence (Admin > Author > Restricted Reader > Reader) is required for legacy BI roles.

### Amazon — Evaluating multi-agent systems for explainability and helpfulness with Amazon Bedrock AgentCore

- Company: Amazon (amazon.com)
- Announced: 2026-10-05T15:50:01+00:00
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://aws.amazon.com/blogs/machine-learning/evaluating-multi-agent-systems-for-explainability-and-helpfulness-with-amazon-bedrock-agentcore/
- Record: https://forck.live/items/16056-evaluating-multi-agent-systems-for-explainability-and-helpfulness-with-amazon
- Subject: Bedrock / Nova

Amazon Bedrock AgentCore Evaluations, a fully managed capability of Amazon Bedrock AgentCore, enables teams to assess agent performance across development and production by measuring accuracy, task success, and behavior across multiple quality dimensions. It supports both built-in evaluators for common dimensions like helpfulness and task success, and custom evaluators for domain-specific validation such as explainability and constraint satisfaction. The solution also integrates with Amazon Bedrock Guardrails for safety controls during execution.

### OpenAI — Our approach to EU text provenance rules

- Company: OpenAI (openai.com)
- Announced: 2026-10-05T15:00:00+00:00
- Category: safety-policy-update
- Coverage: 21 outlets
- Announcement: no
- Group: covered
- Source: https://openai.com/index/eu-text-provenance
- Record: https://forck.live/items/16051-our-approach-to-eu-text-provenance-rules
- Subject: GPT / ChatGPT / API

OpenAI announced its phased approach to text watermarking in response to the EU AI Act. Starting today, API customers globally can opt in to text watermarking for select models, and over the coming weeks an invisible watermark will be added to eligible ChatGPT and Codex text output in the European Union. The company also opened applications for access to its text watermark detector, initially limited to approved researchers and expert organizations.

### NVIDIA — From Scan to Treatment Plan, AI Helps Close Breast Cancer’s Deadliest Gaps

- Company: NVIDIA (nvidia.com)
- Announced: 2026-10-05T13:00:57+00:00
- Category: not stated
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://blogs.nvidia.com/blog/ai-breast-cancer-startups/
- Record: https://forck.live/items/16030-from-scan-to-treatment-plan-ai-helps-close-breast-cancer-s-deadliest-gaps
- Subject: AI platform

Breast cancer is the most commonly diagnosed cancer among American women — yet the gaps in care are wide. A majority of women over age 40 skip the recommended annual screening. Radiologists are reading more mammograms with fewer colleagues. And when a diagnosis arrives, the tests that inform treatment can take weeks to return results. Companies in the NVIDIA Inception program for startups are building AI applications to support clinicians at each of these friction points, including imaging, risk assessment and treatment planning. About 40 million mammograms are performed in the U.S. each year, but a projected shortfall of tens of thousands of radiologists over the next decade is straining the system’s capacity to read them. At the other end of the care timeline, treatment decisions often hinge on genomic assays sent off to outside labs — assays that take weeks to process, at a moment when speed and certainty matter most. The companies below are addressing both ends of that gap, and everything in between, accelerated by NVIDIA AI infrastructure. Automated Imaging Delivers AI-Powered Insights …

### OpenAI — Building advertising for the way people use AI

- Company: OpenAI (openai.com)
- Announced: 2026-10-05T10:00:00+00:00
- Category: capability-change
- Coverage: 1 outlet
- Announcement: yes
- Group: covered
- Source: https://openai.com/index/new-chatgpt-ads-format-and-measurement
- Record: https://forck.live/items/16017-building-advertising-for-the-way-people-use-ai
- Subject: GPT / ChatGPT / API

OpenAI introduced a new visual ad format in ChatGPT, initially testing it during image generation in the US with a select group of advertisers. The company also expanded measurement tools and partnerships, including integrations with Hightouch, Tealium, and LiveRamp, and support for attribution partners such as AppsFlyer and Triple Whale. Early partner findings showed lower cost per acquisition for WeightWatchers and significant lift for wellness brand Dose.

### Cognition — Memory and dreaming: how Devin learns from working with you

- Company: Cognition (cognition.com)
- Announced: 2026-10-05
- Category: capability-change
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://devin.ai/blog/memory-and-dreaming
- Record: https://forck.live/items/16145-memory-and-dreaming-how-devin-learns-from-working-with-you
- Subject: Devin / SWE

Cognition introduced Memory and Dreaming for its AI software engineer Devin. Memory allows Devin to retain user preferences, corrections, and project lessons across sessions, stored in a personal Git repository. Dreaming is a daily background process that deduplicates, links, and refines these memories to improve future sessions.

### Perplexity — 9 RAG Examples: Real-World Use Cases and Architectures

- Company: Perplexity (perplexity.ai)
- Announced: 2026-10-05
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://www.perplexity.ai/hub/blog/rag-example
- Record: https://forck.live/items/16094-9-rag-examples-real-world-use-cases-and-architectures
- Subject: Perplexity

Perplexity's article describes retrieval-augmented generation (RAG) and its common implementations, including hybrid RAG, agentic RAG, graph RAG, multimodal RAG, and corrective RAG. It provides real-world examples such as web search, customer support chatbots, and personal assistants, explaining how each use case applies RAG architecture.

### Suno — 5 Albums You Can't Miss

- Company: Suno (suno.com)
- Announced: 2026-10-05
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://suno.com/blog/albums-to-discover
- Record: https://forck.live/items/16076-5-albums-you-can-t-miss
- Subject: Music / Studio

Suno launched Albums, a feature that allows users to compile their best music into full-length projects. The blog post highlights five albums created by users, including naenia by Oakwood///diskrot, Old Soul by Isaiah Wallace, phenomenal by KakerMix///diskrot, Lost In a Dream by Dream Relic, and VOID by ECHLO.

### Apple — Negotiating Ontological Boundaries in User-Authored Personal Sensing Systems

- Company: Apple (apple.com)
- Announced: 2026-10-05
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://machinelearning.apple.com/research/ontological-boundary-negotiation
- Record: https://forck.live/items/16063-negotiating-ontological-boundaries-in-user-authored-personal-sensing-systems
- Subject: Machine Learning Research

Apple researchers designed two open-ended probes using a Wizard of Oz technique to let people train personalized machine learning systems on phenomena they define themselves. In a week-long study, they identified four sites where ontological boundaries were negotiated: the boundaries of a phenomenon, the subject as part of relations, what counts as signal versus noise, and the objectivity of data. The paper offers starting points for supporting boundary negotiation through design and discusses open-ended probes as a method for ontological design.

### Together AI — Together Link: open models in the harness you already use. Start with one command today.

- Company: Together AI (together.ai)
- Announced: 2026-10-05
- Category: developer-tool-release
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://www.together.ai/blog/together-link-frontier-quality-open-models-in-the-harness-you-already-use
- Record: https://forck.live/items/16057-together-link-open-models-in-the-harness-you-already-use-start-with-one
- Subject: Inference platform

Together AI released Together Link, a tool that connects existing coding agents to open models on its platform, claiming over 50% cost savings. It supports Claude Code, Claude Desktop, Codex, OpenCode, and Pi, with an Auto routing feature that selects models per session. Billing runs on existing Together API keys with serverless pay-as-you-go or credit packs.

### Anthropic — How Cresta turned CX expertise into an agent builder on the Claude Agent SDK

- Company: Anthropic (anthropic.com)
- Announced: 2026-10-05
- Category: not stated
- Coverage: not counted
- Announcement: no
- Group: routine
- Source: https://claude.com/blog/how-cresta-turned-cx-expertise-into-an-agent-builder-on-the-claude-agent-sdk
- Record: https://forck.live/items/16036-how-cresta-turned-cx-expertise-into-an-agent-builder-on-the-claude-agent-sdk
- Subject: Claude

Cresta built Conductor, a natural language agent builder that helps teams create production-ready customer experience agents. Conductor uses the Claude Agent SDK as a general-purpose harness to gather context, use tools, write and run code, and adapt based on results. Cresta reports that in early use cases, Conductor cut initial deployment time roughly in half.

### Cohere — North 2: Enterprise AI without compromises

- Company: Cohere (cohere.com)
- Announced: 2026-10-05
- Category: not stated
- Coverage: 7 outlets
- Announcement: yes
- Group: covered
- Source: https://cohere.com/blog/introducing-north-2
- Record: https://forck.live/items/16029-north-2-enterprise-ai-without-compromises
- Subject: Command

North 2 is Cohere’s biggest upgrade yet — unifying enterprise‑grade security, intelligence, cost governance, and full‑stack control into a single agentic AI platform that runs wherever your data lives. Key takeaways New agent harness: North 2 has a redesigned orchestration system enabling more sophisticated, reliable, multi-step automations and agents across the enterprise. Increased intelligence: Build smarter workflows and richer agent chat experiences using memory, libraries, skills, artifact, and app creation. Model agnostic: Whether you’re using Cohere’s models — or bringing your own — use the models that support your team’s best work. AI sovereignty: You’ll get full control over where and how AI runs, with flexible cloud, on‑premises, and air‑gapped deployments. Enterprise governance: North admin adds granular cost controls, rate limits, user quotas, and org-wide caps for predictable, governed AI operations. For the past few years, enterprises have been racing to deploy agentic AI, and they’ve paid for it in complexity, cost, and compromise. …