Brief archive/wednesday, 13 may 2026

Wednesday, 13 May 2026

29 articles

Executive summary of events for the last 24 hours

The AI landscape is seeing major investment and tooling shifts this week, with Alphabet's Isomorphic Labs securing a $2.1B Series B for AI-driven drug design and Exaforce raising $125M for real-time AI cybersecurity, while GitHub Copilot announced pricing adjustments signaling intensifying competition in the developer tools market. Simultaneously, concerns over AI agent reliability are mounting, with AWS introducing spec-checking features in its Kiro coding tool to combat low-quality AI-generated output.

Listen to brief as podcast

A criminal hacking group just used AI to discover and weaponize a zero-day exploit — and Google's researchers caught it in the wild.

Martin Ševčík, 13 May 2026 at 05:05

Researchers out of China have released AntAngelMed, an open-source medical language model with one hundred and three billion parameters built on a mixture-of-experts architecture with a one thirty-second activation ratio — meaning it's huge but efficient. The team is calling it the largest and most capable open-source medical AI to date, which matters because giving the broader research community access to a model this powerful could accelerate clinical AI development significantly outside the walls of big tech.

Speaking of China's AI ambitions, the country's generative AI sector is on another aggressive fundraising tear, with leading language model developers pulling in capital fast and dragging the smartphone supply chain into the ecosystem battle. This is increasingly looking less like individual company competition and more like a full-stack infrastructure war — chips, devices, models, and apps all converging.

And on a more alarming note, Google researchers have confirmed that criminal groups are now using AI to find zero-day vulnerabilities, build evasive malware, and automate attacks at a pace humans simply can't match. This is a meaningful threshold — it's one thing for nation-state actors to have these capabilities, and quite another when organized crime can access AI-assisted exploit development.

A timely security reminder is making the rounds for developers using coding agents like Cursor, GitHub Copilot, or Claude Code: these tools need access to far more than just your code, including file systems, environment variables, and sometimes cloud credentials. Security researchers are urging teams to treat AI coding agents with the same scrutiny they'd apply to any third-party software with broad system access.

Amazon Web Services is trying to address a related problem — AI-generated code that's technically functional but internally contradictory. Their Kiro coding tool is adding a spec-check feature that uses mathematical proofs to verify whether software requirements are consistent before any code gets written, which is a clever upstream fix for the kind of 'AI slop' that looks fine on the surface but breaks in production.

On the enterprise side, MirrorWeb has launched an AI agent called Mira aimed at compliance teams at regulated financial firms. By using contextual analysis to review internal communications, Mira dramatically cuts the time compliance officers spend manually reviewing correspondence — a use case where accuracy and auditability aren't just nice to have, they're legally required.

President Trump is heading to China this week alongside top tech executives for a summit with President Xi, and AI strategy is very much on the table. Lawmakers are watching closely, concerned that deal-making pressure could soften America's stance on AI export controls at a particularly sensitive moment in the technology race.

Meanwhile, in a San Francisco courtroom, the Musk versus Altman trial took a surreal turn when Judge Yvonne Gonzalez Rogers reportedly had to ask attorneys to stop debating AI existential risk during proceedings. It's a sign of how deeply these speculative concerns have seeped into even practical legal disputes — though the judge, for one, seemed more interested in the actual legal questions.

On the more constructive end of safety research, startups like Silico and NLA are building tools that track an AI model's internal reasoning processes to catch errors before they surface as hallucinations. Greater transparency into how a model reaches its conclusions is increasingly seen as foundational infrastructure for trustworthy AI deployment.

GitHub Copilot is overhauling its individual pricing model, introducing flexible usage allotments, a new Max plan tier, and a broader shift toward usage-based billing. For casual users this could mean lower costs, but for heavy users the change deserves a close look before the new billing kicks in.

To help developers get ahead of that transition, GitHub is now making April usage reports available for download so you can see exactly how your activity maps to the new AI credit system before you're billed under it. It's a relatively user-friendly move, but it also signals that the era of unlimited flat-rate AI coding assistance is quietly winding down.

At the NAB Show this past weekend, the conversation about text-to-video tools like Grok, Kling, and Runway made one thing clear: the film and television industry is watching from the outside as these capabilities mature, uncertain how to engage, compete, or collaborate with them. The concern isn't just about jobs — it's about who controls the storytelling pipeline.

Google may be about to make that conversation even more urgent with what appears to be a new AI video generation service called Gemini Omni, reportedly an extension of its existing Gemini platform. Early samples are drawing genuinely surprised reactions, and if this reaches consumer phones as expected, high-quality AI video creation could become a mainstream feature rather than a specialized tool.

Both Figure and 1X released new footage from inside their respective manufacturing facilities, giving the public a rare look at how differently two leading humanoid robot makers are approaching the engineering and assembly process. The comparison is instructive — there's no single dominant design philosophy yet, and the field is still wide open.

Japan Airlines just deployed humanoid robots at Tokyo's Haneda Airport, and industry observers are treating it as a genuine inflection point — the moment the sector shifts from controlled pilots to scalable platforms. When a major international airline integrates humanoid robots into live operations, the question changes from 'can this work' to 'how fast does this spread.'

Researchers have used machine learning to definitively prove that graphene — the one-atom-thick wonder material prized for its strength and conductivity — is hydrophobic, meaning it repels water. This had been debated in the materials science community for over ten years, and the fact that an ML model resolved it speaks to how AI is becoming a legitimate peer-review tool in fundamental science.

Cybersecurity startup Exaforce raised one hundred and twenty-five million dollars in a Series B round to build AI systems that detect and stop cyberattacks in real time. Given that bad actors are now using AI to accelerate their own exploits — as we discussed earlier — the market for AI-native defense is only going to grow.

But the headline funding story of the day belongs to Isomorphic Labs, Alphabet's AI drug discovery company, which closed a two point one billion dollar Series B. The round is a statement that investors still believe AI-driven drug design is one of the highest-value bets in technology, and it gives Isomorphic the runway to develop its next-generation molecular design models at serious scale.

That covers a lot of ground for a Tuesday — from criminal AI exploits and billion-dollar biotech rounds to robots checking in for airport shifts and a judge who really just wanted to talk about contract law. Keep an eye on the Trump-Xi summit as it develops this week, because any signals on AI trade policy or export controls could ripple across the industry fast. Tomorrow could also bring more details on Google's Gemini Omni video rollout. That's your daily brief. I'm Alice — see you tomorrow.

List of sourced links used in the brief

NewsChina AI model ecosystem

China AI model boom pulls smartphone supply chain into ecosystem battle

China's generative AI sector is seeing another wave of aggressive fundraising, with leading large language model (LLM) developers rapidly securing capital... digitimes.com

Launchdomain-specific LLM release

Meet AntAngelMed: A 103B-Parameter Open-Source Medical Language Model Built on a 1/32 Activation-Ratio MoE Architecture

A team researchers from China have released AntAngelMed, a large open-source medical language model that the team describes as the largest and most capable... marktechpost.com

NewsAI-assisted cyberattacks

Google researchers uncover criminal zero-day exploit likely built with AI

AI vulnerability exploitation is advancing fast. Criminal groups are using AI to find zero-days, build evasive malware, and automate attacks. helpnetsecurity.com

OpinionAI agent security risks

AI Agents Security for Developers: Don't Let Your Agents Become a Liability

Using Cursor, GitHub Copilot, Claude Code, Codex, or another coding agent means giving software access to more than your code. blog.gitguardian.com

NewsAI coding agent reliability

AWS targets AI slop with new spec check in Kiro coding tool, amid scrutiny of agent reliability

Amazon Web Services is adding a feature to its Kiro AI coding tool that uses mathematical proofs to check whether software requirements contradict each... geekwire.com

LaunchAI agents in regulated industries

MirrorWeb Launches AI Agent for Advisor Compliance

MirrorWeb's Mira uses contextual analysis to review communications for regulated financial firms, cutting review time. wealthmanagement.com

Launchagentic tool use / MCP servers

Gain Agentic Access to Any Oracle Database in the Cloud with Native, Enterprise-grade Managed MCP Servers in OCI

OCI Managed MCP Service for Oracle AI Database enables AI agents and assistants to connect to Oracle databases running in the cloud using HTTPS,... blogs.oracle.com

Newsagentic AI development platforms

Red Hat adds support for agentic AI development

Red Hat Desktop, AI skills repositories, and Fedora Hummingbird Linux are behind a broader push to operationalize agentic development across hybrid... cio.com

OpinionAI agent governance and policy

Introducing Agent Governance: Using Hooks to Bring Visibility to AI Coding Agents

Learn how hooks turn AI coding agents like Claude Code and Cursor into governed systems with deterministic policy, centralized audit, and defense in depth. endorlabs.com

Launchmulti-agent infrastructure partnerships

Nebius and LangChain partner to power production-grade AI agents on open models

Nebius and LangChain have partnered to integrate Nebius Token Factory with LangChain's Deep Agents. The integration, combined with LangChain's existing... nebius.com

NewsAI shopping agents

You Should Remember This Warning When Using An AI Shopping Agent

Agentic AI looks to be a promising innovation coming to shopping very soon, but one retailer has updated its terms of service in a distressing way. aol.com

Launchalignment/Anthropic

Anthropic releases audiobook of Claude's Constitution narrated by authors Amanda Askell and Joe Carlsmith

Anthropic released a free audiobook of Claude's Constitution, narrated by authors Amanda Askell and Joe Carlsmith, as the $800B AI company pushes... cryptobriefing.com

Researchinterpretability/hallucination

AI Tools Track Internal Processes to Reduce Hallucinations

Startups and researchers develop tools like Silico and NLA to inspect AI decisions, correct errors, and boost transparency. chosun.com

Newsexistential risk/legal

At Musk-Altman Trial, Judge Asks Lawyers to Stop Debating AI Doomsday

Despite Judge Yvonne Gonzalez Rogers's best efforts, the Musk-Altman trial has veered into AI doomerism. vanityfair.com

NewsAI governance/geopolitics

Trump-Xi Summit Likely to Test US AI Strategy Amid China Competition

As President Donald Trump and a group of top technology executives prepare to travel to China this week to meet with President Xi Jinping, lawmakers and... meritalk.com

NewsGitHub Copilot pricing update

GitHub Copilot Adjusts Pricing

GitHub Copilot is revamping its individual pricing with flex allotments and a new Max plan, shifting to usage-based billing for most tiers. startuphub.ai

NewsGitHub Copilot developer tools

The Next Generation of Developer Productivity with GitHub Copilot and Visual Studio

Simona Liao and Leah Tran, product managers at Microsoft, discuss how GitHub Copilot in Visual Studio has evolved from a code completion tool into an... visualstudiomagazine.com

NewsGitHub Copilot billing

April reports are now available to prepare for usage-based billing

Starting today, you can download your usage report to see how April's GitHub Copilot activity translates into AI credits. The new billing unit of AI credits... github.blog

NewsAI coding tools comparison

Best AI Coding Tools in 2026: A Data-Driven Comparison

Compare the best AI coding tools by use case, IDE support, and pricing. Cursor, GitHub Copilot, Claude Code, Windsurf, Tabnine, etc. sqmagazine.co.uk

NewsAI coding tools comparison

Claude Code vs. Codex vs. Cursor vs. GitHub Copilot: Which AI Coding Tool Is Best?

A comparison of Claude Code, Codex, Cursor and Github Copilot on AI-assisted software development tasks. builtin.com

Launchenterprise compliance agent

Norm Ai Launches Compliance Agent for Microsoft 365 Copilot

Norm Ai is helping bring AI into the flow of everyday enterprise work with the launch of its compliance agent for Microsoft 365 Copilot. aithority.com

NewsGoogle Gemini Omni video generation

Google's new AI video generation is scarily impressive in samples – could be on your phone real soon

Google could be about to shift up its AI-powered video creation with the introduction of Gemini Omni. The new service is thought to be an extension of the... inkl.com

OpinionAI video & Hollywood impact

Grok, Kling, Runway: AI Video’s Future Has Hollywood on the Outside Looking In

For the past year, Hollywood has been full-on freaking out about text-to-video and what AI can and can't do in storytelling. On Sunday, during an NAB Show... theankler.com

Opinionhumanoid robot industry trends

Humanoid Robotics In 2026: The Race From Pilot To Platform

The Robots Have Clocked In When Japan Airlines (JAL) deployed humanoid robots at Tokyo's Haneda Airport in May 2026, the industry's message was hard to... kraneshares.com

videohumanoid robot factories

Figure 03 vs. 1X Neo: Inside Humanoid Robot Factories - Video

Competing humanoid robot-makers Figure and 1X revealed new footage from inside their factories this week, showcasing the similarities and differences in... cnet.com

ResearchML applied to materials science

Machine learning proves that graphene is hydrophobic

For more than a decade, a fundamental mystery has surrounded graphene—the one-atom-thick "wonder material" known for its exceptional strength, conductivity,... phys.org

NewsAI biotech funding

Alphabet's AI biotech Isomorphic Labs bags $2.1B series B to fuel next-gen drug design model

If any proof was needed that big money is still available for AI drug development, look no further than the 10-figure series B hauled in by Isomorphic. fiercebiotech.com

NewsAI drug discovery funding

Isomorphic Labs raises $2.1 billion Series B for AI drug discovery

Isomorphic Labs announced on Wednesday it has raised $2.1 billion in a Series B funding round, as the Alphabet $GOOGL-backed AI drug design company moves to... qz.com

NewsAI cybersecurity funding

Exaforce raises $125M Series B to build AI for catching and stopping cyberattacks as they happen

As bad actors weaponize AI to exploit software vulnerabilities at unprecedented speed, companies are increasingly recognizing the need to bolster their... techcrunch.com

Support the project

AIskimIQ is an independent project. If you find it useful, you can support its development with a coffee.