The Informed Layer · Newsscan
Daily briefing

Opus 5.5 debuts; US lawmakers push frontier AI rules

← All issues

Anthropic / Claude ecosystem

Claude Code v2.1.280 makes Opus 5.5 the default, including on $20 Pro and Team Standard plans

The Claude Code 2.1.280 update installs Opus 5.5 as the default Opus model with 1M-token context, and switches the default from Sonnet to Opus on the $20 Pro and Team Standard tiers. Updated cost accounting and logging behaviour accompany the change.

Ten Claude Opus 5.5 agents find faster shortest-path algorithm and formally prove it

Vals AI reports that ten Claude Opus 5.5 agents discovered C-HD, a shortest-path algorithm with an asymptotic improvement over Dijkstra for graphs in a specific density range, and produced a formal correctness proof in Lean within 15 hours.

Anthropic and OpenEvidence partner to bring medical AI to 100+ low- and middle-income countries

Anthropic and OpenEvidence have announced a partnership to deliver AI-powered clinical decision support grounded in medical evidence to more than 100 low- and middle-income countries. The deal aims to widen access to evidence-based care in regions with scarce clinical resources.

Anthropic launches Claude Opus 5.5, first of a new 5.5 model family

Anthropic has released Claude Opus 5.5, the first model in a new Claude 5.5 family, reporting strong benchmark results at lower running costs than Opus 5. The launch lands amid a broader frontier push toward cheaper high-performance models.

Frontier model providers

Google DeepMind adds secure persistent memory to Private AI Compute

Google DeepMind introduced a secure, persistent server-side memory layer for Private AI Compute that preserves device-held keys and end-to-end encryption. The design enables cross-device personal-assistant continuity without exposing raw user data to the cloud provider.

OpenAI opens its Daybreak cyber defence programme to Ukraine

OpenAI is offering free access to its Daybreak cyber defence programme to Ukraine to help protect civilian infrastructure amid ongoing cyber threats.

OpenAI launches GPT-6 Sol and GPT-6 Luna with 50% lower API prices

OpenAI announced GPT-6 Sol and GPT-6 Luna, faster and more affordable additions to the GPT-6 family, with API prices cut by 50% and availability across the API, Codex and ChatGPT. The company claims higher accuracy at lower cost per token, with expanded enterprise-tier access.

Austrian Academy and Mistral release Apollo, an Ancient Greek LLM trained on 600 million words

The Austrian Academy of Sciences and Mistral have released Apollo, described as the first advanced LLM for Ancient Greek, trained on 600 million words. It is already being used to reconstruct damaged papyrus fragments for classical scholars.

AI developer tooling & infrastructure

Notta launches MCP server and CLI to pipe meeting context into AI agents

Transcription provider Notta introduced Notta MCP and a companion CLI that give AI agents standardised, direct access to meeting content without manual exports, streamlining integration into agent workflows.

JetBrains unveils Air, an open system for agentic software development

JetBrains has unveiled Air, an open system of tools for managing AI-powered software delivery workflows across human developers and autonomous agents. The system aims to orchestrate agentic development within existing IDE-centred toolchains.

Datris releases open-source data control plane for AI agents

Datris has released an open-source data control plane for managing AI agents' data workflows, covering production data handling and recovery for agent systems.

Cloud & platform providers

AWS Security Hub AI Inventory adds Azure self-hosted instance support

AWS extended Security Hub's AI Inventory to discover and catalog AI assets on Azure self-hosted instances, giving security teams cross-cloud visibility of AI workloads and their security posture.

Amazon launches CloudWatch Omni, AI-first observability for agents and applications

Amazon CloudWatch Omni is now generally available as an AI-first observability experience that integrates OpenTelemetry with CloudWatch across multi-account and multi-cloud environments, aimed at monitoring agents and modern applications.

AI policy, regulation & governance

UK parliament invites OpenAI, Meta, Anthropic and Google DeepMind to AI hearing

UK lawmakers have invited the major frontier AI developers to give evidence on safety, pre-release testing and incident reporting ahead of potential regulatory changes, with the hearing set for 13 October.

Newsom names expert panel to deliver California AI executive order, including kill-switch work

California Governor Gavin Newsom announced a group of world-leading experts charged with delivering his AI executive order, including advancing the creation of a frontier-model kill switch under the state's independent AI safety process.

Sanders and Casar introduce bill to ban artificial superintelligence and pause advanced AI

Bernie Sanders and Greg Casar introduced legislation to create a new federal agency that would ban artificial superintelligence and pause advanced AI development until safety regulations and oversight are established.

Welch and Bennet propose independent federal agency to regulate frontier AI

Senators Peter Welch and Michael Bennet released a proposal for an AI Regulator Act establishing an independent federal agency to regulate frontier AI models and large social-media platforms, aimed at preventing catastrophic AI risk and protecting consumers.

DeepSeek set to brief UN Security Council on AI risks

Chinese frontier lab DeepSeek is set to brief the UN Security Council on the security implications of advanced AI, entering high-level global governance discussions. It follows news that Sam Altman will brief the Council next week.

Cloudflare and Microsoft dismantle AI-powered MFA phishing ring targeting Australia

Cloudflare and Microsoft took part in a coordinated international operation disrupting EvilTokens, a phishing-as-a-service platform behind AI-assisted MFA-bypass attacks that targeted Australian organisations among others, via takedowns and domain seizures.

UK health regulators commit to joint guidance on AI regulation

UK health and care regulators, including the General Pharmaceutical Council, have signed a joint statement of intent committing to coordinated, high-level AI regulation principles to maintain public trust in healthcare AI use.

Anthropic and OpenAI urge Australia to relax ban on training AI models on local content

Anthropic and OpenAI have called on Australia to relax its effective ban on training AI models on Australian content, proposing a narrow conditional approval path that couples training access with creator protections. The push comes weeks after Anthropic paused Australian model training over unresolved copyright rules.

Australia has released exposure draft legislation for a Digital Duty of Care that would overhaul the online safety regime with proactive, systems-based obligations for online services spanning content, platform design and AI-related risks.

Industry & market moves

TD Bank commits $25 million to strategic AI relationship with Cohere

TD Bank Group announced a strategic relationship with Cohere and its Layer 6 research team, committing up to $25 million over three years to accelerate responsible AI adoption and practical banking use cases in Canada.

Bird.com secures $450 million to launch communications infrastructure for AI agents

Bird.com has secured $450 million in financing led by J.P. Morgan as it launches communications infrastructure for AI agents and rolls out its Agentic Harness platform in the US.

Accenture invests in Within and forms partnership to accelerate enterprise AI deployment

Accenture announced a strategic investment in and new partnership with Within, with the combined offering aiming to take clients from process discovery to AI agent deployment at enterprise scale.

TEKEVER raises $580 million Series D at $6.4 billion valuation

European autonomous-systems company TEKEVER closed the first close of a US$580 million Series D led by UC Investments and Baillie Gifford, reaching a US$6.4 billion valuation as a leading European AI autonomous-systems firm.

Adobe completes acquisition of AI enhancement firm Topaz Labs

Adobe has completed its acquisition of Topaz Labs, bringing the Emmy Award-winning AI image and video enhancement technology into Adobe's creative ecosystem and Creative Cloud workflows.

Realset AI and Flatkey raise $10 million Series A for real-world training data

Realset AI and Flatkey raised a $10 million Series A to build real-world training data for frontier models and embodied agents, expanding real-workplace data capture and open benchmarks.

Upwind acquires Aegis and launches dedicated AI Security Labs

Cloud security firm Upwind has acquired Aegis and formed Upwind AI Security Labs, a dedicated research unit focused on securing AI agents, skills and plugins.

Aleph Alpha and Cohere pursue business combination to offer sovereign AI through Cohere

Law firm Milbank confirmed it is advising LEA Venturepartner on a business combination of Germany's Aleph Alpha and Canada's Cohere. The combination would unite two enterprise AI providers to offer sovereign deployments through Cohere.

Mistral acquires adtech startup Pimento in EUR 12.7 million deal

Filings show Mistral has acquired adtech startup Pimento in a EUR 12.7 million cash-and-shares deal, its third acquisition of the year as it expands its commercial AI offering.

Brahma AI raises $150 million round led by Multiples

Brahma AI has raised a $150 million funding round led by private-equity firm Multiples to broaden its global expansion and R&D.

OpenAI acquires experimentation platform Statsig for $1.1 billion

OpenAI has acquired Statsig, a product experimentation startup, in a $1.1 billion deal intended to accelerate OpenAI's own AI-enabled software development.

AI product & feature launches

Bell and Cohere deploy Canadian-built cybersecurity AI for threat investigation

Bell Canada has deployed a domain-specific cybersecurity AI model built with Cohere to speed threat investigations inside Bell Cyber's operations, framed as a homegrown sovereign security capability.

Accenture Edge and AWS launch mid-market AI collaboration with six marketplace offerings

Accenture Edge and AWS announced a collaboration delivering six ready-to-deploy, AI-enabled and secure cloud offerings for mid-market organisations through AWS Marketplace.

Guava launches Daytona voice model with open voice-agent benchmark

Guava released Daytona, a state-of-the-art voice model, alongside the open Guava Voice Index benchmark for evaluating voice-agent performance against humans.

Light Origins launches Light-O1 for cross-embodiment robot skill transfer

Light Origins launched Light-O1, showing that cross-embodiment transfer for real-world robot tasks improves as human-action pretraining scales, alongside a preview model and released code.

Simular launches Sai and SaiFleet for autonomous computer fleets

Simular released Sai, an autonomous computer-use agent, and SaiFleet, enabling parallel fleets of agents to operate Windows, macOS and Linux machines at low cost, pitching scalable digital labour.

Black Forest Labs debuts FLUX 3 Action, an open-weights robotics model topping RoboLab-120

Black Forest Labs released FLUX 3 Action, an open-weights AI robotics model that tops the RoboLab-120 leaderboard at roughly half the size of competitors, with faster inference.

Research with immediate practical relevance

DeepSeek details DSec, its production-scale sandbox infrastructure for agent training

A new DeepSeek paper describes DSec (DeepSeek Elastic Compute), a production-scale sandbox platform for training and evaluating autonomous agents, built around multi-tenant sandboxes and high throughput.

Windows malware uses four AI models voting on its next move

Researchers detailed CLOSEDQUORUM, Windows malware that queries up to four AI services and acts on their votes when deciding next steps, including credential and crypto-wallet theft - an early case of malware outsourcing decisions to LLMs.