Dakarda AI Newsletter · 5 June 2026
Friday Edition
Microsoft Scout changes the AI game
Microsoft has unveiled Scout – an autonomous agent that works in the background of M365, doesn't wait for a prompt, and has its own identity. Plus a leak of a new Anthropic model, the first regulatory moves by the US and EU, and a financial agent that replaces junior analysts. Friday will be busy.
Intro · Alex
This week ends with a strong hit from Microsoft, which at Build 2026 presented Scout – an autonomous agent working non-stop in the background of the entire Microsoft 365. This is not another Copilot – Scout manages tasks in Teams, Outlook, OneDrive, and SharePoint on its own, has its own identity in Entra ID, and works asynchronously. You don't have to tell it what to do – it senses the context and acts on its own. This is the moment when 'agent' ceases to be a buzzword and becomes a real work tool. In today's edition, I also reviewed a leak of a new Anthropic model called Claude Oceanus (possible successor to Mythos), the first regulatory moves by the US and EU towards oversight of frontier models, an autonomous financial agent from AlphaSense, and two important texts from the world of cybersecurity. Buckle up – the weekend with an accelerated AI loop is coming.
What's worth knowing
Microsoft Scout – the first Autopilot for M365 that works always and everywhere
Scout is a breakthrough compared to Copilot – it doesn't wait for a prompt, but constantly monitors context in Teams, Outlook, and SharePoint, executing tasks asynchronously. Built on the OpenClaw framework, with its own identity in Entra ID. The Frontier early access program is already launching.
Leak of Claude Oceanus – new Anthropic model with 1-million context? Premiere reportedly on June 16
In the Claude console, the identifier claude-oceanus-v1-p appeared – likely the successor to Mythos, offering 1M token context and speed ~52 tok/s. If rumors are confirmed, competition in the frontier LLM segment will intensify before the summer.
US and EU launch frontier AI regulations in the same week – voluntary review and 60-member scientific panel
The European Commission appointed 60 independent experts to classify GPAI models, and Trump signed an EO introducing voluntary review of frontier models. Two continents, one direction: model safety is no longer an option, it becomes a standard.
From the tech world
AI-Driven Exploitation is destroying vulnerability management – how to defend?
The window between vulnerability disclosure and exploitation has shrunk from days to hours. Case study: a team used Claude Mythos Preview to find over 10,000 critical vulnerabilities in a month. A practical wake-up call for every SOC.
Over 700 Ghost CMS sites infected with fake Captcha (ClickFix)
Attackers used SQL injection (CVE-2026-26980) in Ghost CMS, stole Admin API keys, and injected malicious JavaScript into articles. Users encountered a fake Captcha. Full attack chain from exploit to malware distribution.
Tip of the day
Agent-as-an-analyst – how to use AI for autonomous research
AlphaSense has showcased SuperAnalyst – an agent that 24/7 searches financial data, aggregates reports, and generates analytical summaries. This is a model you can replicate in your own area: define sources, build a multi-step prompt, set a schedule, and let the agent work in the background. In practice: choose a domain with a high volume of daily reports (markets, competition, regulations), integrate the agent with a content acquisition tool (RSS, API), and set up a daily briefing. Start with simple flows in n8n or LangChain – you don't need to build a production system right away.
Tool of the issue
AlphaSense SuperAnalyst – autonomous agent that replaces a junior analyst 24/7
SuperAnalyst independently searches, aggregates, and analyzes financial data, reports, and news, generating analytical summaries without human involvement. It works around the clock, executes multi-step research flows, and monitors markets in real time.
Reading list
AI-Driven Exploitation is Destroying Vulnerability Management
The Hacker News analyzes how AI has shortened the response time to vulnerabilities – and why traditional patching is no longer enough. A team used Claude Mythos to detect over 10,000 critical flaws in a month.
Over 700 Ghost CMS sites infected with fake Captcha (ClickFix)
Sekurak describes a campaign using SQL injection in Ghost CMS – from theft of Admin API keys to injection of malicious JavaScript. A great case for analysis for admins and pentesters.
An always-on agent is not a bug – it's a feature of the future. Scout, SuperAnalyst, Oceanus – this week, the concept of 'act, don't ask' won. I wonder how much longer the boundary between 'tool' and 'colleague' will exist. Let me know what you think – I reply to every email.
Want the next issue in your inbox?
Subscribe — every issue delivered directly to you.