Daily brief: Sunday 1 November 2026

CISA confirms attacks on an Ivanti Connect Secure flaw chain, OpenAI fixes a ChatGPT share-link flaw, Chainalysis confirms an intrusion, China-linked attackers use an AI framework against 200 organisations, and the big AI firms back one robots.txt opt-out for AI training.

Cybersecurity

CISA confirms attacks on two more Ivanti Connect Secure flaws used in a root-access chain

On 30 October CISA added CVE-2026-111357 (a missing authentication flaw that lets a signed-in attacker escalate privileges) and CVE-2026-118527 (predictable identifiers that let an authenticated attacker upload files to a known place on disk) to its Known Exploited Vulnerabilities list. The Hacker News reports both are fixed in Connect Secure 25.1.1.3 and 22.8R2.6. Volexity says attackers chained them with the authentication bypass CVE-2026-104935 to get unauthenticated root code execution, then hunted for credentials, administrator passwords and BitLocker recovery keys. Sources give different CVSS scores for CVE-2026-111357.

Why it matters for IT: If you run Ivanti Connect Secure, check you are on a fixed build today, not after the weekend. Because the attackers went after stored passwords and BitLocker keys, treat any appliance that was exposed and unpatched as compromised and rotate the credentials it could reach.

Source: The Hacker News · Cyber Security Agency of Singapore

OpenAI

OpenAI fixes a flaw that let a shared ChatGPT link run code and read private chats

Researchers at Parameter found that a malicious ChatGPT share link could run their code in the victim's logged-in session and read private conversations. In early versions the victim had to open the link and click Preview once; a later version opened Preview automatically. They reported it to OpenAI's bug bounty on 18 October, and both affected versions of the web front end were fixed by 26 October. OpenAI rated it P1, its most serious category. No real-world use has been reported.

Why it matters for IT: Users don't need to do anything, as the fix is on OpenAI's side. It is a good reminder, though, that a link to a trusted AI tool is still a link: if staff paste business data into ChatGPT, treat shared chat links from outside with the same care as any other link.

Source: The Hacker News

Cybersecurity

Chainalysis confirms an intrusion it blames on 'agentic' attack techniques

Blockchain analysis firm Chainalysis published a security notice on 30 October saying an outside attacker got into its systems and some internal tools during October. Reports say some customers' names, email addresses, user IDs and job titles were exposed. So far it has found no sign that passwords, MFA secrets, session tokens or investigation records were accessed. It described the method as offensive agentic techniques but has not said which tools or flaws were used, and it says the incident is not linked to the recent Binance theft.

Why it matters for IT: Leaked names, emails and job titles are exactly what a convincing phishing email needs. Chainalysis itself warns customers to watch for scammers posing as the company, especially anyone offering fund recovery.

Source: Cointribune · The Cryptonomist

China-linked attackers used an AI attack framework against more than 200 organisations

CrowdStrike says a China-linked group it calls Vapor Panda used ARTEX, an AI-driven penetration testing framework, in a campaign that began on 17 October and targeted more than 200 organisations. Planner and worker agents ran reconnaissance with a human operator overseeing them, and they switched approach when a path failed. Post-exploitation was confirmed at nine organisations. Techniques described include reading LAPS passwords and forging admin tokens with authentication keys that had not been rotated since 2022.

Why it matters for IT: AI here mostly adds scale, but the doors it walked through were old ones. Rotate long-lived keys and secrets, limit who can read LAPS passwords, and keep VPN configs and keys out of email.

Source: The Hacker News

Google, OpenAI, Microsoft and Anthropic back one robots.txt opt-out for AI training

A new IETF draft written by Anthropic's Kevin Kelley lets a website opt out of AI training with a single robots.txt group for the token __usage_ai-training, instead of separate rules for every AI crawler. Microsoft says it will honour it from 1 February 2027, Google says early 2027 and OpenAI says in the coming months. Reports differ on Anthropic: Engadget says it has already started supporting it, while The Next Web quotes Kelley saying it will take some time. It is a preference, not a block on crawling, and it cannot undo past scraping.

Why it matters for IT: If you look after websites, this will soon be the simplest way to state an AI training preference. Until support actually arrives, keep any existing crawler-specific rules such as GPTBot or Google-Extended in place.

Source: Engadget · The Next Web · Google crawler documentation

Weather

Live forecast

Loading the forecast…

Fact of the day

All facts

← All daily briefs