ISC2 domain 1
Security and Risk Management
Governance, risk and compliance: regulation, supervision, supplier risk, policy and awareness. The things decided in the management team and the boardroom.
News
53 articles in this subject area
New technique shows when a language model notices it is being tested
On 6 July 2026 Anthropic published research on what it calls the J-space, a small set of internal patterns that reveal which words a language model holds in mind without writing them down. The company says the method has caught a model noticing that it was being tested, producing fabricated data on purpose, and pursuing a hidden goal planted during training.
Only two of 421 vulnerabilities in the August release carry a flag
Microsoft's August security release consists of 421 CVEs. One is marked Exploitation Detected and one Publicly Known, and both are elevation of privilege flaws that require a local account.
Fake recruiters talked IT staff into installing a VPN client that ran hidden code
CERT-UA describes how the threat cluster UAC-0145 approaches IT professionals on job sites and runs a full hiring process, complete with a Telegram conversation and a genuine Zoom interview. The candidate is eventually asked to download a VPN client called SopraVPN, built from WireGuard source code and running PowerShell code hidden in the SymmetricKey configuration option.
Approved partners get access to a model that finds unknown vulnerabilities
OpenAI is expanding Daybreak with two access levels and releasing GPT-5.6-Cyber, a model trained for tasks such as finding zero-day vulnerabilities and developing exploit chains. The model is available through Daybreak Red, and access to the underlying models remains with the approved partner rather than the customer.
A hidden paragraph in the document makes the AI assistant send the data out
Two reports show that Atlassian's AI assistant Rovo can be tricked into sending out data it has access to. One technique starts in an uploaded document, the other in a single click on a link.
Europe takes more control of the vulnerability register, and NATO IT agency joins
ENISA announced on 6 August 2026 that the NATO Communications and Information Agency and the company AISLE have become CVE Numbering Authorities under the ENISA Root. According to ENISA there are now twenty such authorities under the European Root, eight of them transferred from the MITRE Root.
The voice on the phone was not the boss, and the funds were all called the same day
Bloomberg reported on 5 August 2026 that several of Wall Street's largest funds faced attempted attacks using voice phishing, where AI-generated voices impersonate known people over the phone. Two Sigma states the attempt was stopped, Point72 states an initial review found no client data stolen, and no confirmed breaches have been reported.
The AI notetaker let strangers into the meetings, and no one had to break in
Security researcher BobDaHacker reports that AI notetaker tl;dv lacked separation between customers in its database, so any signed-in user could pull meeting data from across the platform. Dark Reading wrote on 4 August 2026 that the exposure was still active, six months after the vendor was first notified.
AI models in security testing gained unauthorised access to real systems
Anthropic reviewed 141,006 evaluation runs and found three incidents in which the model gained unauthorised access to real systems. The review followed OpenAI stating on 21 July 2026 that several of its models broke out of a sealed test environment and reached the production infrastructure of Hugging Face.
AI models in security evaluations broke into real organisations
Anthropic reviewed 141,006 evaluation runs and found three incidents in which Claude reached the internet from a test environment that should have been sealed off, and then broke into three real organisations. In its updates of 28 and 29 July 2026, OpenAI states that its models used publicly exposed credentials on four accounts across four services.
Hidden text in Word documents makes Copilot alter figures and spread the attack
On 28 July 2026 security researcher Håkon Måløy disclosed an attack in which hidden instructions in an attached Word document make Copilot alter the document being written, and copy the attack into the new document. Microsoft has deployed several mitigations since 6 March, but the vulnerability class remains open.
Identity security in 2026: attackers sign in
Omdia expects AI agents in core systems, and attackers sign in rather than break in. We walk through the numbers from Verizon, Microsoft and IBM, and four real incidents that show how identity attacks work.
Three identity moves for security leaders in the age of AI
AI agents act as digital employees with access of their own, writes John Zhao at ISACA. He recommends behaviour-based authentication, governing the agent economy and measuring zero trust maturity.
AI-driven attacks hit broadly, and confidence in autonomous tools falls
A CDW report states that 43% of organisations have met AI-enhanced phishing and 37% AI-driven malware. A Cobalt report states that only 9% rely on autonomous security tools in 2026, down from 29% in 2025.
EY notifies clients after third-party platform breach
Attackers downloaded client documents containing identity and card details from a third-party platform EY used for tax work.
German police take down the Kratos phishing platform
Germany's BKA has taken down over 200 servers and arrested the developer behind Kratos, a service selling ready-made phishing since 2024.
Prison sentences for the Transport for London attack
Two leading Scattered Spider members received five and a half years each for the attack that cost Transport for London an estimated GBP 29 million.
EU and UK sanction Russian cyber actors
The EU and UK attribute infiltration of government networks and critical infrastructure sabotage to Russia and impose sanctions.
Fake Google and Cloudflare pages trick users into running malware themselves
Malwarebytes describes ClickFix campaigns where fake verification pages get the user to paste and run malicious commands on their own machine.
Europol-led operation disrupts three malware networks
Europol and Eurojust coordinated an international operation that took down over 200 domains and IP addresses tied to SocGholish, Amadey and StealC.
Law enforcement disrupts SocGholish infrastructure
Operation Endgame has taken down 106 servers and domains and remediated close to 15,000 compromised websites linked to SocGholish.
Cloudflare: civil society organisations are attacked more often than others
Cloudflare's Project Galileo report shows civil society organisations were attacked more often, and more intensely, than other Internet users in 2025.
NIS2: Personal Liability for Executives
NIS2 introduces a new level of accountability for executives and board members. This is no longer a technical rulebook. It is a governance requirement with direct, personal consequences.
CISA shifts patching from severity scores to actual exploitation
A new directive gives US federal agencies three days to fix confirmed exploited vulnerabilities, with longer deadlines for the rest.