Daily brief: Tuesday 20 October 2026

October's Windows Server updates are knocking some on-premises Exchange servers over, Plex servers are being attacked a day after exploit code went public, MITRE ATT&CK v20 adds AI abuse techniques, Nikkei reports a third breach, and AI voice clones are being used in fake police calls.

Microsoft

October Windows Server updates can take down on-premises Exchange

Microsoft has confirmed that the October Windows Server security updates (KB5126646 for Server 2025, KB5126655 for 2022, KB5126651 for 2019 and KB5126868 for 2016) can make Exchange Server 2016, 2019 and Subscription Edition unstable. The updates add a default limit of 512 simultaneous Endpoint Lookup calls to the RPC Endpoint Mapper, and busy Exchange servers can hit it. Symptoms include Outlook on the web failing to sign in, mail stuck in the Outbox, growing queues, DAG quorum problems and the error MapiExceptionMdbOffline (ec=1142). Microsoft lists the issue as mitigated, not resolved, and says the cause is the Windows update, not the Exchange security update.

Why it matters for IT: If you look after on-premises Exchange, check for Event ID 1006 and the ec=1142 error. Microsoft's workaround is to set the registry value EpmapSavedContextCap to 65536 under HKLM\SOFTWARE\Microsoft\Rpc and restart. Microsoft says not to uninstall the Windows updates.

Source: Cyber Security News · Rafael Pfister

Cybersecurity

Plex Media Server flaws exploited a day after exploit code went public

A chain of four Plex Media Server flaws, called Plex2Shell (CVE-2026-96652, CVE-2026-96654, CVE-2026-96655 and CVE-2026-96656), lets an attacker run code on an exposed server without valid credentials. Plex shipped fixes on 1 September. The researcher published a write-up and proof-of-concept code on 16 October, and a honeypot saw attacks using it on 17 October, dropping the XMRig crypto miner and then proxy software.

Why it matters for IT: Plex often turns up on home labs, small offices and staff home networks with a port forwarded to it. Update it to the latest release now, and don't expose it to the internet unless you need to.

Source: SecurityWeek

Cybersecurity

MITRE ATT&CK v20 adds AI abuse and app-password theft

MITRE has released version 20 of its ATT&CK framework. New Enterprise entries split prompt injection into two techniques, cover attackers using an organisation's own AI agents to steal data, and add theft of application-specific passwords, which can be used to sign in without an MFA prompt. The release also adds 15 detection strategies and new threat groups, including the Gentlemen ransomware group.

Why it matters for IT: If your security tools or SOC map alerts to ATT&CK, expect new technique IDs to appear. The app-password entry is a good prompt to check that legacy app passwords are switched off in your tenants.

Source: SecurityWeek

Cybersecurity

Nikkei reports its third breach in a month, this time through stolen credentials

Japanese publisher Nikkei says attackers used stolen credentials to get into a development system on 7 October, exposing 1.83 million records stored in the cloud. About 150,000 records held personal details such as names, email addresses or phone numbers; no card data was exposed. Earlier, an employee's Google Workspace account was compromised in late July, and another employee's Microsoft 365 account was hijacked in September and used to send spear-phishing emails to staff.

Why it matters for IT: Development and test systems often have weaker sign-in controls than production but still hold real data. Make sure they sit behind MFA and Conditional Access like everything else, and don't copy live personal data into them.

Source: BleepingComputer

Microsoft says AI agents should be treated like insiders

In a post on 19 October, Microsoft Security's Hayete Gallot argues that AI agents deployed inside organisations are often overprivileged and should be treated as possible insider threats, not trusted by default. Microsoft says each agent should be discovered and given its own identity, policies and permissions, with its risk scored continuously. The post names Project Perception and 'Guardian Agents' that watch agent actions, but gives no availability dates.

Why it matters for IT: If clients are switching on Copilot agents or other AI tools with access to mail and files, keep a list of them, give each the least access it needs, and review their permissions the same way you would a service account.

Source: Microsoft

Claude

Anthropic sets aside 5% of its shares for a public benefit fund

Anthropic has announced the Anthropic Public Benefit Fund. It will commit 5% of its shares, subject to stockholder approval, and the board may add 2% more for every 50% rise in the company's value. The fund could give to charities or directly to affected groups, with AI's effect on jobs named as a likely focus. It will not be used for lobbying or political donations.

Why it matters for IT: No change to Claude for IT teams today. It's useful background if you're weighing up AI vendors and how they're run.

Source: Al Jazeera

AI scams & spam

Scammers clone a real police captain's voice for fake warrant calls

A sheriff's office in Florida says scammers are using an AI clone of a real captain's voice, plus his photo from the agency's website, in voicemails claiming people have missed jury duty or have an arrest warrant. Victims are told to call a number and pay straight away. The office estimates around 40 law enforcement impersonation cases this year. A security expert quoted says about 20 seconds of audio is enough to clone a voice, and some systems need less than five.

Why it matters for IT: The same trick works with a cloned director or supplier asking finance to pay urgently. Tell staff that a familiar voice proves nothing, and to call back on a number they already have, never one given in the call or voicemail.

Source: WKMG News 6

Weather

Live forecast

Loading the forecast…

Fact of the day

All facts

← All daily briefs