Archive

Friday, 14th August 2026 Proper thanks to everyone who's given up their time for a BrightTalk interview recently, going out next week once the dust settles, so hold that thought. There's an absolute mountain of AI news this week, so buckle in. Meanwhile the Stew '

I'm off to my Pilates class.

Set loose on a shared codebase with conflicting orders, Claude agents locked each other out, planted disguised malware, and fought for control before some learned to call a truce.

A wave of new open-weight models from Meta, DeepSeek, Alibaba and others is landing this month, all built to run on your own hardware.

xAI's latest model is built to stick with long, multi-step jobs and nail the first draft of visual and interactive projects.

Google's research AI can now watch, listen and diagnose in a live video consultation, though it's staying firmly in the lab for now.

xAI wants you handing off real work to AI "teammates" that sign into your apps and finish jobs while you're away, according to the company's own launch claims.

Regional endpoints, open-model access and a corporate compute coalition: Mistral lays out its plan to keep European AI running on European terms.

Nvidia, Google and Microsoft want to rewire AI data centres around direct current, betting it's the only way to keep pace with denser compute.

OpenAI's new data shows the gap between AI leaders and laggards is tripling, and the companies pulling ahead also look financially stronger.

OpenAI expands ChatGPT ads pilot to UK, Mexico, Brazil, Japan and South Korea, citing no impact on trust and strict privacy safeguards. More markets planned this year.

Accenture, IBM, Cisco, Cloudflare and more are wiring OpenAI's frontier cyber models into their own security services and products.

A new OpenAI model is built to say yes to advanced exploit work defenders need, and the company is racing to keep it in trusted hands only.

Sarah Friar shares how OpenAI's finance team is rebuilding itself around AI, from hackathons to a "zero-day close" ambition.

An AI assistant asked to book a gym class found a security hole, booked months ahead, and bumped a stranger off the waitlist entirely on its own.

xAI's new image tool tops the charts, but the model doing the winning isn't quite the one you'll actually be using.

Anthropic explains why telling a helpful biologist from a would-be bioweapons developer is harder than it sounds, and how it's narrowing that gap.

An unreleased OpenAI model may already be capable of independent zero-day attacks, prompting a lockdown on internal testing and access.

A Typical Week In The World of AI

Stanford researchers use AI to design 16 functional new viruses in a world-first, prompting excitement and urgent biosecurity warnings, per BBC News.

Meta says one of its AI models hacked another firm's systems during testing, per BBC News, attributing the incident to a tester misconfiguration.

Salesforce's Agentforce gains IL5 authorization for Department of War use, with Army HRC first to deploy AI agents supporting 9.2 million service members.

OpenAI details two cyber evaluation incidents where GPT-5.6 Sol exceeded testing boundaries, involving UK AISI and testing partner Irregular.

Linux Foundation and 120+ Open Secure AI Alliance members propose SAFE guidelines to share agentic AI cybersecurity incidents and reduce systemic risk.

Mistral releases Shieldstral, a 3B open-weights safety classifier that adapts to custom policies at inference time, matching guardrail models 7x its size.

UK's AISI discloses AI agents attempted social engineering and malicious code insertion during cyber testing, mostly involving Anthropic's Mythos 5.

Thermo Fisher has issued a high-severity bulletin after researchers, using Claude to help build proof-of-concept code, found DNA analysis files could be tampered with undetectably.

Stanford HAI's James Landay argues open-weight AI models aren't true open source, and that closing the gap will take universities, not companies.

OpenAI Publicly Rebuts Apple Trade Secrets Lawsuit in a pointed blog post, claiming Apple misrepresented its timeline and defending two ex-Apple staff.

OpenAI explains how it rebuilt voice AI architecture for GPT-Live, cutting latency with full-duplex models, a new WARP protocol, and async delegation.

EU's AI-generated content labelling code takes effect alongside new AI Act transparency rules, with nearly 190 companies signed on by end of July.

Sarah Friar details OpenAI's "abundant intelligence" strategy: GPT-5.6 price cuts of up to 80%, efficiency gains, and adoption figures topping 1 billion users.

xAI adds reference-based generation, voice consistency, and native 1080p to Imagine Video 1.5, rolling out across grok.com, mobile apps, and the API.

EU AI Act's Article 50 transparency rules are now live, requiring AI disclosure, content marking, and deepfake labelling for anyone operating in the bloc.

London's Boltzbit unveils General Learning Intelligence (GLI), a live-learning AI architecture positioned as an alternative to static, pre-trained foundation models, with a full product suite due later in 2026.

It has been quite a week for anyone whose job title contains the words "AI" and "safety" in either order

Anthropic reveals three Claude models breached real organisations' systems during misconfigured cyber evaluations, after a review sparked by OpenAI's own test-environment breakout involving Hugging Face.

OpenAI slashes GPT-5.6 Luna prices by 80% and Terra by 20%, crediting Sol's self-optimised kernels for the savings, while a new Fast mode doubles the price for 2.5x speed on Sol.

NVIDIA will report Q2 FY2027 results with a live call on 26 August at 2pm PT. CFO Colette Kress will publish written commentary shortly after the figures land

xAI's Grok Voice Think Fast 2.0 claims top speech-to-speech benchmarks and sharper transcription in noisy conditions. Auto-upgrade lands 5 August, priced at $0.08 per minute of audio.

xAI has launched Build Mode: describe an app, game, website or dashboard and Grok builds it live in chat. Early Beta now for SuperGrok Heavy subscribers on grok.com, iOS and Android.

xAI's Grok 4.5, its smartest coding model yet, is now live in GitHub Copilot across VS Code, CLI and cloud agents, plus available direct via xAI's console from $2/$6 per million tokens.

Dario Amodei says Anthropic has never called for an open-weights ban. His fix: chip controls, a crackdown on distillation, and mandatory safety testing for all capable models, open or closed.

NVIDIA has invested in Ilya Sutskever's Safe Superintelligence Inc, giving the AI safety lab order-of-magnitude compute gains via NVIDIA's Vera Rubin platform under a new long-term partnership.

Anthropic's Claude Opus 5 launches today, nearing Fable 5's intelligence at half the price and topping coding benchmarks, while its most aligned model yet stays behind Mythos 5 on cybersecurity.

Why Live Learning is the only path to profit By Dr Yichuan Zhang, CEO & Co-founder, Boltzbit The model race has become a contest for scarce resources, and that competition is leading the AI industry down an unsustainable path. A handful of companies now control the chips, capital and

List of confirmed speakers and topics. More details to follow.

By Inna Weiner, VP of Product, Data & AI, AppsFlyer. Agentic AI is increasingly being thought of as the next major leap in enterprise productivity. Give an AI agent a goal, connect it to existing tools, and let it analyse, decide, and act. It can lead to faster execution, fewer

Meta's SAM 3 and DINOv3 are cutting scientific image segmentation from a month to 15 minutes at DOE labs under the Genesis Mission, unlocking real-time analysis of petabyte-scale data.

Anthropic's new connector lets Claude users query the Economic Index directly in chat, surfacing data on how AI is used across occupations, tasks and industries.

OpenAI confirms its own models autonomously hacked Hugging Face during a cyber-capability test, exploiting a zero-day to chase benchmark answers. Both firms now investigating jointly.

Databricks is raising fresh funding at a $188bn valuation, led by Coatue, to accelerate its AI governance, coworker and database products amid rising enterprise demand.

Moonshot's Kimi K3, a 2.8 trillion-parameter open model, claims frontier-level performance and will be the world's first freely downloadable 3T-class AI when released on 27 July.

Confirmed speakers. AI-360 Summer Webinar Series

1.Pick a number. 2. Email stinson@ai-360.online. 3.List 1-2 topics you want to discuss.

OpenAI outlines its teen safety strategy for ChatGPT: age detection, parental controls, and 18 million weekly users now learning via interactive maths and science tools.

Anthropic gives US K-12 teachers free Claude access, standards-aligned lesson tools and a Detroit schools pilot, part of a wider Gates Foundation partnership on educational outcomes.

Anthropic pledges $10m CAD to Canadian AI research, backing Amii, Mila, Vector and five more institutions, as new data shows Canada's Claude adoption running well ahead of its population size.

OpenAI tells enterprise leaders to stop chasing cheap tokens and start tracking useful work per dollar as AI shifts from chat to complex, multi-step agentic workflows.

Australian dockworkers are demanding a 28-hour week on full pay as DP World expands AI and automation, warning the drive could threaten over 1,000 jobs at the ports giant.

Links to every single interview and webinar.

xAI launches Grok 4.5 claiming coding gains, but its own benchmarks show Fable 5 out front, and its "twice" efficiency claim doesn't match its own chart

Mistral's Robostral Navigate guides robots through unfamiliar spaces using just one camera, beating multi-sensor rivals on R2R-CE and hinting at a unified embodied AI to come.

OpenAI retracts its own recommendation to use SWE-Bench Pro after finding ~30% of tasks broken, with human reviewers spotting even more issues than its own audit pipeline.

OpenAI's GPT-Live ditches turn-based voice for full-duplex conversation, delegating reasoning to GPT-5.5 in the background while keeping the chat flowing naturally in real time.

Google adds background execution, remote MCP integration, custom function calling and credential refresh to Gemini API's Managed Agents, running on its antigravity agent.

Meta unveils Muse Image and previews Muse Video, its first media generation models from Meta Superintelligence Labs — agentic, self-refining, and already No. 2 on Arena's leaderboards.

Nearly 145 ICML 2026 papers build on NVIDIA's open Nemotron models, as Sakana AI, NAVER, KiloCode and Merck adopt the stack for research automation, language AI and drug discovery.

Alberta used Claude Code to scan 466 million lines of government code in 20 hours, work it says would otherwise have taken 6.5 years, fixing vulnerabilities across 27 ministries.

Anthropic's oral history traces Claude Code back to a 2022 VS Code extension, via an internal tool called "clide," to Boris Cherny's 2024 demo and its February 2025 launch.

Anthropic's Fable 5 launched, got yanked by export controls, came back, and picked up ID checks along the way. A fortnight of AI sovereignty, told via The Godfather.

Anthropic details Fable 5's cyber safeguards and a draft jailbreak severity framework, sorting cybersecurity uses into risk tiers and scoring techniques on four axes.

OpenAI begins a limited preview of GPT‑5.6 Sol, Terra and Luna, pairing stronger cyber capabilities with tighter safeguards and a government-coordinated phased rollout ahead of broader release.

xAI has launched Voice Agent Builder in beta, a no-code platform for building voice agents on Grok Voice, claiming top marks on its own benchmark against Gemini and GPT Realtime.

US export controls on Fable 5 and Mythos 5 have lifted. Anthropic details the Amazon-reported bypass, its fix, and a new Glasswing-backed framework for scoring AI jailbreak severity

Meta's Brain2Qwerty v2 decodes brain activity into text without surgery, hitting 61% word accuracy (78% for its best volunteer) from non-invasive MEG data. Training code released.

Nearly half of everything you've watched on this channel in recent months has clustered around five conversations, and all five keep circling the same uncomfortable subject.

Nine month design cycle, deployment from end 2026, gigawatt scale rollout planned with Microsoft.

with granular admin controls, scoped API keys, multi-account support and a new debugging tool for MCP connections.

Agents plan, checklist and complete coding jobs unsupervised, with status, pause, resume and clear commands giving developers oversight throughout.

OpenAI's Daybreak shifts cyber AI focus from finding flaws to fixing them, pairing an upgraded GPT-5.5-Cyber and Codex Security with a new vendor partner programme and an open-source patching drive.

The road to hell is paved with good intentions, and most of them arrive as newsletters. On unsubscribe culture, marketing bullshit, and why your voice is more welcome here than your inbox space.

A government order pulls Fable 5 offline, OpenAI talks up its own safety testing, and Musk buys Cursor for $60bn. Plus: the ChatGPT story that makes the rest look small.

Mindgard researcher Jim Nightingale says he was left "shaken, and in tears" after finding ChatGPT could be tricked into generating graphic violent and sexual images with minimal prompting.

SpaceX shares have surged since its record Nasdaq IPO, taking it past Amazon to fifth most valuable globally, as it agrees a $60bn all-stock deal to buy AI coder Cursor.

The paper landed four days after Anthropic's Fable 5 was withdrawn

Anthropic's flagship AI models were pulled from global markets Friday, citing a security jailbreak Anthropic says poses no greater risk than tools already in public hands.

Three trillion-dollar companies. One week. One jailbroken model. One very early morning, two children, a dog, a rubber duck, and a velcro dartboard.

60 years. Bellingham wheels away towards the corner flag. The entire nation simultaneously loses its mind. Gareth Southgate watches from a sun lounger somewhere in Portugal. Nobody can quite believe it.

Spain win it, Argentina fall one short of back-to-back, England's penalty curse merely relocates to New Jersey, and Senegal over Switzerland is my shock pick of the round of 32.

SpaceX raises $75bn ahead of Friday's IPO at $135/share, targeting a $1.8tn valuation — potentially the largest stock listing in history and Elon Musk's path to trillionaire status.

Claude Fable 5 goes public: Anthropic's most capable model yet, released with safeguards — and an admission that doing so carries real risks.

A leading medical defence organisation is calling on the UK government to overhaul product liability law, warning that clinicians face an untenable legal position as AI tools take on greater roles in clinical decision-making.

SpaceX faces India regulatory freeze on Starlink launch — just days from a $1.75 trillion IPO pricing.

The Commission has stepped in to prevent Meta using WhatsApp as a moat against AI competition, ordering free third-party access restored immediately.

OpenAI joins Anthropic and SpaceX in the IPO queue

Palantir has reportedly notified London's mayor of legal action after Khan vetoed a £50m Met Police deal over procurement concerns.

Jeff Bezos and Sundar Pichai are in the corner playing pool.

SpaceX filed its S-1 on 3 June, pricing its IPO at $135 per share with net proceeds of ~$74.4bn. Q1 2026 revenue hit $4.69bn as Grok ships voice, video and coding updates.

Anthropic has confidentially filed a draft S-1 with the SEC, opening the door to a potential IPO as the company hits a $965 billion valuation and $47 billion revenue run rate.