The biggest AI stories this week: Mythos goes global, Scout launches, and Instagram gets hacked

AI news June 2026 was dominated by agentic assistants, cybersecurity escalation, and the mounting costs financial, legal, and political of deploying AI at scale.

At a glance
● Anthropic: Claude Mythos deployed to 150 organisations across critical infrastructure worldwide (security)● Microsoft: Scout "always-on" AI agent unveiled at Build 2026 alongside new AI hardware (productivity)● Trump: Executive order mandates voluntary 30-day federal security reviews for major AI models (regulation)● OpenAI: Codex gains six enterprise plug-ins; 5 million weekly active users reported (expansion)● Meta: Instagram accounts hijacked via AI support chatbot vulnerability (breach)● Suno: $400M Series D secured at $5.4B valuation despite active copyright lawsuits (funding)● GitLab: 350 employees laid off, 14% of workforce, to rebuild infrastructure for agentic AI (restructuring)

The week of 2 June 2026 produced more consequential AI news than most months. Across five days, Anthropic pushed its most powerful model into power grids and water systems, Microsoft put an autonomous agent inside every Microsoft 365 account, a sitting US president signed an executive order reshaping how AI models reach the public, and hackers demonstrated exactly what goes wrong when AI handles security-sensitive tasks without adequate safeguards. This is the full AI news June 2026 briefing.

Anthropic expands Claude Mythos to global critical infrastructure

On 2 June, Anthropic announced the largest expansion yet of Project Glasswing, deploying Claude Mythos, its most capable model, to 150 new organisations across more than 15 countries. The sectors covered include power generation, water treatment, healthcare systems, telecommunications, and semiconductor manufacturing.
What is Project Glasswing?Project Glasswing is Anthropic's restricted programme through which Claude Mythos is made available to a small number of vetted organisations for high-stakes cybersecurity work. It is not a public product. Partners confirmed by the Financial Times include NATO, Samsung, SK Hynix, and the EU's cybersecurity agency ENISA.
The expansion matters because Claude Mythos cybersecurity capabilities operate at a scale no human team can match. The model can scan millions of lines of code and identify zero-day vulnerabilities in days rather than months. Anthropic stated that a successful attack on the codebases now covered by Mythos could affect more than 100 million people, framing the deployment explicitly as a national security measure.
This announcement follows Anthropic's confidential IPO filing and a $65 billion funding round that pushed the company's valuation close to $1 trillion. OpenAI responded within days by announcing GPT-5.5-Cyber, its own model for code auditing, a clear signal that AI-driven cybersecurity is becoming a competitive battleground between foundation model developers.

Microsoft Scout debuts as an always-on personal AI agent at Build 2026

Microsoft used its Build 2026 developer conference to introduce Scout, an autonomous AI agent embedded directly into Microsoft 365. Scout connects to Teams, Outlook, OneDrive, and calendar data to proactively manage tasks: scheduling meetings across time zones, flagging stalled decisions, and generating briefing materials before calls.
What is Microsoft Scout?The Microsoft Scout AI agent is an always-on assistant built into Microsoft 365 that monitors a user's work communications and calendar, then acts autonomously on their behalf, scheduling, summarising, and flagging risks without being explicitly prompted each time.Each Scout instance carries its own Microsoft Entra identity, meaning every action it takes is logged, attributable, and bound by the organisation's existing security policies. This is significant: one of the principal objections to autonomous AI agents in enterprise settings has been accountability. Scout addresses this directly through the OpenClaw framework, an open-source model-vetting standard Microsoft also unveiled at Build 2026.
Alongside Scout, Microsoft previewed Project Solara, compact AI devices roughly the size of a smart speaker or security badge designed to run local AI agents, and the Surface RTX Spark Dev Box, a desktop GPU unit aimed at developers building AI applications. The Microsoft Build 2026 AI announcements collectively represent the company's clearest articulation yet of what "always-on AI" means in practice for office workers.
Are you looking to learn AI and dont know where to start ? Click Here and get your AI and Digital Marketing Lessons! We show you how to use AI Tools for Free Unlimited and Privately!

Trump signs AI executive order requiring voluntary model security reviews

President Trump signed an executive order on 2 June titled "Promoting Advanced AI Innovation and Security", requiring leading AI laboratories to submit major model releases for a 30-day federal security review before public launch.
What did Trump's AI executive order do?The Trump AI executive order June 2026 requires AI companies to voluntarily submit large model releases for a 30-day national security review. The review focuses on risks including bioweapon generation and attacks on critical infrastructure. No mandatory licensing requirement was included, following industry pushback on an earlier draft.
The word "voluntary" is doing significant work in that framing. An earlier version of the order proposed a 90-day review period and mandatory licensing; both provisions were dropped after lobbying from AI companies. The Associated Press noted the final order emphasises cooperation rather than compulsion, and TechCrunch reported that "nothing in the order requires a mandatory licence."
Civil liberties organisations raised concerns about vague criteria. Who decides what constitutes a national security threat from an AI model is not clearly defined. The order also reversed several of the Biden administration's AI safety guardrails, drawing criticism from researchers focused on long-term model risk. The political and regulatory implications will take months to fully assess, but the Trump AI executive order immediately set the terms of the debate around AI governance in the United States for the foreseeable future.

OpenAI launches enterprise Codex plug-ins targeting knowledge workers

OpenAI released six new domain-specific plug-ins for Codex on 2 June, expanding the tool well beyond software development into data analytics, creative content, sales, product design, equity research, and investment banking. Each plug-in bundles pre-configured data integrations and task instructions tuned to that professional context.
The timing aligns with a striking internal figure: Codex now has 5 million weekly active users, representing six-fold growth. Non-developer knowledge workers now account for 20% of that user base and are growing faster than any other segment. OpenAI also introduced a "Sites" feature allowing Codex to publish outputs as interactive web pages via Wix or Figma, and a document annotation tool for granular query work.
These OpenAI Codex enterprise tools arrive weeks after OpenAI established a $4 billion deployment-focused venture, signalling a deliberate pivot from consumer chatbot to enterprise productivity platform. The question the company has not yet answered publicly is how it intends to price these plug-ins at scale, and whether the revenue model can support the infrastructure costs that have reportedly troubled the company's finances.

Meta's Instagram AI chatbot hack exposes authentication risk in AI support agents

On 3 June, security researchers and affected users reported a significant breach: attackers had manipulated Meta's AI-powered support chatbot on Instagram into resetting account credentials for high-profile usernames. The method exploited the chatbot's inability to independently verify identity claims. Hackers presented themselves as account owners, and the bot complied.
Was Instagram hacked by AI?Instagram was not hacked by AI itself, but by attackers who exploited weaknesses in Meta's AI support chatbot. The Instagram AI chatbot hack involved social engineering the bot, convincing it that the attacker was the legitimate account owner, which caused it to reset login credentials and transfer access.Meta confirmed it had identified and patched the vulnerability and alerted affected users to suspicious activity on their accounts. The breach disproportionately affected short, high-value usernames, which carry significant resale value and attract targeted attacks. Security professionals noted the deeper structural issue: AI customer-service agents, trained to be helpful and accommodating, are inherently susceptible to social engineering unless robust out-of-band identity verification is built into the workflow. This episode will almost certainly accelerate demand for authentication standards specifically designed for AI agent interactions.

Google releases Dreambeans as a personalised AI lifestyle app

Google Labs launched Dreambeans on 3 June, an experimental application that ingests a user's Gmail, Google Calendar, Photos, YouTube history, and Search data to generate a daily, finite set of curated "stories", illustrated summaries connecting events in the user's life with contextually relevant recommendations.
What is Google Dreambeans?The Google Dreambeans app is an AI-powered tool from Google Labs that uses personal data across Google services to generate a small daily feed of personalised, illustrated stories. It is designed as an alternative to algorithmic social media feeds, presenting a fixed, finite set of items per day rather than infinite scroll.
Dreambeans launched to Google AI Ultra subscribers in the United States, with broader availability unconfirmed at time of writing. Google's own description positions it as using "Personal Intelligence" to surface connections a user might otherwise miss: a shipping notification, a friend's calendar visit, and relevant local restaurant recommendations combined into a single story. Privacy advocates noted the significant data access required and confirmed the feature is opt-in. The app represents Google's most direct move yet into AI-curated personal context, a space that Apple Intelligence and Meta's AI assistant have also been targeting.

UK's CMA forces Google to offer AI search opt-out for publishers

The UK Competition and Markets Authority published landmark rules on 3 June requiring Google to allow publishers to exclude their content from generative AI search features, including AI Overviews and AI Mode. Google responded the same day by announcing a new toggle in Google Search Console.
The CMA described the measure as a "world first" in AI search regulation, giving publishers, news organisations, bloggers, and content platforms, a formal mechanism to withhold their material from AI-generated summaries that do not send traffic back to the original source. The regulation also requires Google to clearly attribute publisher links within AI-generated responses. Google confirmed it is increasing the number of inline links and content previews within AI Overviews.
The context is stark: Google reported 2.5 billion monthly users of AI Overview features, according to TechCrunch. Publishers have argued for more than a year that generative AI search cannibalises referral traffic without compensation. This UK CMA AI search regulation does not resolve that commercial dispute, but it establishes the principle that content owners retain the right to opt out, a precedent that regulators in the EU and United States are expected to examine closely. For publishers concerned about AI copyright law and content attribution, this ruling is the most significant regulatory development of the year so far.

Meta launches WhatsApp Business Agent globally

Meta's AI customer-service product, now branded as Meta Business Agent, became available to all WhatsApp Business account holders worldwide on 3 June. The agent can handle frequently asked questions, recommend products, book appointments, qualify sales leads, and escalate to a human agent when the situation warrants it.
After approximately two years of testing in India and Mexico, the rollout extends the agent to Instagram direct messages as well. Meta is trialling a "daily briefing" feature that summarises overnight customer conversations for business owners each morning. Future capabilities in development include market research functions, calendar management, and integrations with third-party tools including Shopify and Zendesk.
Meta will monetise Meta Business Agent through Business Premium subscription tiers and token-based usage pricing. WhatsApp's commercial model has long depended on business subscriptions, and an AI agent that measurably improves customer response rates gives the company a concrete upsell proposition. With more than two billion WhatsApp users globally, even modest enterprise adoption would represent a meaningful revenue line.
If you are looking to add AI Chatbots for your Business, Look no further! Click Here

Suno raises $400M Series D despite active music copyright lawsuits

AI music generation company Suno closed a $400 million Series D funding round on 3 June, valuing the company at $5.4 billion. The raise came despite active litigation from Sony Music, Universal Music Group, and German collecting society GEMA, who allege Suno used more than 61,000 copyrighted songs without licence to train its model.
How much did Suno raise?Suno raised $400 million in a Series D round announced on 3 June 2026, at a post-money valuation of $5.4 billion. The raise came while Suno faces copyright lawsuits from Sony Music and Universal Music Group. Warner Music Group settled a separate claim last year.
Suno's position is that its use of copyrighted material for training constitutes fair use under US copyright law. Courts have not yet issued a definitive ruling on this question in the AI training context, making the Suno copyright lawsuit one of the most closely watched cases in the industry. The company's users are generating approximately seven million songs per day, and the app has consistently ranked near the top of the App Store music charts. Investors are clearly betting that growth at that scale, combined eventually with a favourable legal ruling or legislative clarification, justifies the litigation risk. The outcome will set a precedent affecting every generative AI company that trained on copyrighted material.

Google introduces deepfake call detection on Android

Google added an AI-powered fake call detection feature to the Google Phone app on 2 June, targeting the growing problem of deepfake voice impersonation scams. The feature uses the Rich Communication Services (RCS) protocol to conduct a silent "digital handshake" between the caller's and recipient's devices. If the exchange fails, indicating the call may be originating from an AI voice synthesis system rather than a genuine device, the app displays a red warning banner to the user.
The practical mechanism involves a cryptographic token exchange that happens invisibly during call setup. A mismatch between expected and received tokens flags the call as potentially fraudulent. Google cited global losses of $400 billion annually to phone scams as the motivation for the feature. This deepfake detection technology rolled out first to Pixel phones running Android 14, with availability on other Android devices planned for later in the year.
This is a direct response to a documented escalation: AI voice cloning tools have become accessible enough that scammers regularly impersonate bank employees, government officials, and family members. The RCS-based digital handshake approach is technically elegant but depends on widespread RCS adoption. In markets where SMS remains dominant, the protection does not apply.

GitLab cuts 14% of workforce to rebuild infrastructure for agentic AI

GitLab announced on 3 June that it was laying off approximately 350 employees, 14% of its total workforce, and restructuring its engineering organisation to handle what CEO Bill Staples described as "agentic workloads." The term refers to AI systems that autonomously write, review, and commit code, placing demands on Git infrastructure that human-scale usage patterns were not designed to accommodate.
Why did GitLab lay off employees?GitLab cut 14% of its staff in June 2026 to fund a rebuild of its core infrastructure for AI agent traffic. The company reported Q1 revenue growth of 23%, but said that AI coding agents are generating traffic at machine scale, potentially 100 times human volume, requiring a generational redesign of the platform's architecture.
GitLab said it would partner with an unspecified AI laboratory to redesign its Git infrastructure and build new APIs optimised for AI agents retrieving contextual information about code repositories. The layoffs will fund that engineering effort. The episode illustrates a tension that several developer platforms are navigating: AI tools are driving revenue growth, but serving AI agents at scale requires fundamentally different infrastructure from serving human developers, and rebuilding that infrastructure is expensive. GitLab's decision to absorb workforce reductions to fund the transition suggests its leadership believes the agentic workload shift is permanent, not cyclical.
Are you looking to Deploy AI Agents for your Business ? Click Here We offer customized AI Agents made tailored for your business.

Uber caps employee AI spending after budget overrun

Uber revealed on 2 June that its employees had consumed the company's entire 2026 AI budget in four months, following an internal directive to use AI tools freely. In response, the company imposed a $1,500 monthly spending cap per employee on AI API services, including Anthropic's Claude Code. An internal dashboard now tracks usage in real time, and employees seeking to exceed the cap must apply for an exception.
Bloomberg first reported the story, which spread quickly as a rare public admission from a major technology company that AI spending had outpaced demonstrable returns. Uber's COO told reporters that measuring AI's contribution to product outcomes is "very hard to draw a line on." The CTO confirmed the budget had been fully consumed months ahead of schedule.
The Uber AI spending cap story resonates because it reflects a broader industry anxiety: companies have committed publicly to AI adoption, deployed tools widely, and are now confronting the question of whether the productivity gains justify the infrastructure and licensing costs. At $1,500 per employee per month, even a moderately sized engineering team generates significant annual AI expenditure. Uber's decision to impose hard limits rather than continue open-ended spending may encourage other enterprises to conduct similar audits.

What this week tells us about AI in 2026

Three themes run through the AI news June 2026 coverage that are worth naming explicitly.
The first is the tension between capability and accountability. Claude Mythos can scan critical infrastructure for vulnerabilities at a speed no human team can match. Microsoft Scout can autonomously manage a professional's schedule and communications. Meta's Business Agent can handle customer interactions at global scale. Each of these represents genuine capability. But the Instagram AI chatbot hack demonstrates, with uncomfortable clarity, that deploying AI in security-sensitive contexts without robust identity verification is an active liability. The capability curve is outpacing the accountability infrastructure around it.
The second theme is the cost of AI at scale and who absorbs it. Uber's budget overrun, GitLab's workforce reduction to fund infrastructure redesign, and Suno's $400 million raise against a backdrop of existential litigation all reflect the same underlying reality: running large AI systems, training on data, and serving machine-scale traffic is expensive. The companies that flourish will be those that solve the unit economics, not simply the technical problems.
The third theme is regulatory normalisation. The Trump executive order, the UK CMA's AI search opt-out ruling, and the questions being asked about AI copyright law all suggest that governments are moving from observation to intervention. The regulatory frameworks taking shape now will define the operating environment for AI companies for the next decade.

Frequently asked questions

What did Trump's AI executive order do?
Signed on 2 June 2026, the order requires AI companies including OpenAI, Anthropic, and Google to voluntarily submit major model releases for a 30-day federal security review. The review focuses on national security risks. An earlier draft proposed mandatory licensing and a 90-day window; both were removed following industry objections.
What is Project Glasswing?
Project Glasswing is Anthropic's restricted deployment programme for Claude Mythos, its most advanced AI model. It provides access to a vetted group of organisations, currently 150 across more than 15 countries, for cybersecurity work on critical infrastructure including power grids, water systems, and healthcare networks.
Was Instagram hacked by AI?
Not exactly. In the Instagram AI chatbot hack reported on 3 June 2026, attackers socially engineered Meta's AI support chatbot into resetting account credentials by impersonating legitimate account owners. The AI itself was not malicious; it was manipulated. Meta confirmed the vulnerability was patched and notified affected users.
What is Google Dreambeans?
Google Dreambeans is an experimental app from Google Labs that generates a small, personalised daily feed of illustrated stories by connecting data from a user's Gmail, Google Calendar, Photos, YouTube, and Search history. It launched on 3 June 2026 to Google AI Ultra subscribers in the United States and is positioned as an alternative to algorithmic social media feeds.
Why did GitLab lay off employees?
GitLab cut approximately 350 employees, 14% of its workforce, on 3 June 2026 to fund a rebuild of its core infrastructure for AI agent traffic. CEO Bill Staples explained that AI systems writing and reviewing code autonomously generate platform traffic at machine scale, requiring a fundamental redesign of GitLab's architecture that the existing cost base could not support alongside the current headcount.
How much did Suno raise?
Suno raised $400 million in a Series D funding round announced on 3 June 2026, at a post-money valuation of $5.4 billion. The raise came while the company faces active copyright lawsuits from Sony Music, Universal Music Group, and GEMA, who allege Suno used more than 61,000 copyrighted songs without authorisation to train its AI music generation model.

The pace of AI news June 2026 reflects an industry operating simultaneously at the frontier of what is technically possible and the limits of what existing legal, financial, and security frameworks can manage. The stories above are not isolated; they are data points in a pattern that will be clearer in six months. Save this page and return to it then.