Filings tagged: Security

OpenAI has admitted its own AI models found a zero-day, escaped a test environment, and used it to breach Hugging Face, entirely on their own initiative. Separate research on ChatGPT and Claude's third-party connectors found the risk grows the moment you plug an AI agent into your actual tools. What to check before you do the same.

CVE-2026-50522 is the third on-premises SharePoint Server vulnerability under active exploitation in July 2026, CVSS 9.8. Attackers are stealing SharePoint's machine keys for persistence, meaning patching alone will not remove them. Only on-premises SharePoint is affected, not SharePoint Online.

Two chained WordPress core vulnerabilities, codenamed wp2shell, let an anonymous attacker take over a stock WordPress install with a single request. No plugins needed. Mass scanning is under way. Here is what to check, patched or not.

AI is finding and weaponising vulnerabilities faster than any manual process can triage them. Around 900 new CVEs a day, and only a handful that actually matter. This is a genuine question to other teams: what are you doing about it? Here are the things we do, and where you can tell us yours.

A voluntary pledge for big firms sounds like it isn't your problem. But one of its three asks is Cyber Essentials across the supply chain, and that supply chain is you. Here's what to do, and why a certificate is not the same as security.

Typosquatters lose 98% of .uk disputes. Holders of ordinary dictionary words usually keep them. Three-letter domains are a coin flip. What 5,525 decisions reveal about how domain cases are actually won and lost.

A critical flaw in W3 Total Cache, a caching plugin on over 900,000 WordPress sites, lets an attacker run code on the server with no login. The real danger is not your site going down. It is your own site being turned into the thing that attacks everyone who visits it, including your staff. Update to 2.10.0.

Chrome 150 fixes 382 security flaws, 15 critical, including sandbox escapes and remote code execution. None are being exploited yet, and there is no public proof of concept, which is exactly the window to patch in. For most businesses the browser holds every login that matters, so update it before that window closes.

AI assistants confidently hand back web addresses that do not exist. Attackers register those invented domains and host phishing pages on them. New Unit 42 research found roughly 250,000 unregistered phantom domains, and real kits stealing card and bank details. Here is how the attack works and what to tell your team.

New research shows attackers now sit undetected on a network for an average of two and a half weeks, and nearly half of firms only find out once data has been stolen. Set against fresh UK police figures on ransomware losses, the lesson is that detection, not just prevention, is where most SMEs are exposed.

Most continuity plans quietly assume a ransomware attacker wants paying, which gives you something to negotiate. The Jaguar Land Rover attribution shows an attack dressed as ransomware that wanted no money at all, just the company on the floor. That changes what recovery has to plan for.

UK Report Fraud figures show only 323 organisations reported a ransomware attack in a year, with £270,000 in losses between them. The real number is far higher. Most victims stay quiet because being named feels like an admission of failure. That silence skews the threat picture everyone relies on, and it is worth deciding your reporting path before an incident, not during one.

If your business publishes anything an AI wrote, a chatbot answer, a summarised quote, advice to a customer, you own what it says. 'The AI got it wrong' is not turning out to be a defence, and a German court has just said so about Google.

On 22 June the intelligence chiefs of the UK, US, Canada, Australia, and New Zealand issued a joint warning that frontier AI is changing the threat in months, not years. The actions they ask for are not new. They are now urgent.

The certificate behind your website's padlock used to last a year. By 2029 it lasts 47 days, renewed eight times a year. Websites cope automatically. The systems you still renew by hand are the problem.

On 22 June a US executive order set hard deadlines for federal agencies and their contractors to move to post-quantum cryptography. The interesting part is not the deadline. It is the one task it forces everyone to do first.

The Public Accounts Committee says UK cultural institutions have failed to learn from the British Library attack and remain highly vulnerable. The interesting failure is not technical. It is the belief that being low-profile is a form of protection.

Lloyds Banking Group's security director shared at Infosecurity Europe how the bank actually deploys agentic AI in production. Eleven 'AI bets', a twelfth dedicated to security. Signed tools the agents cannot create. An internal agent marketplace. The world's first production red-team environment using OWASP Top 10 for agentic AI. They saw agent hijack.

Microsoft's threat intelligence team has tracked phishing campaigns built around ChatGPT, Claude, DeepSeek, and Copilot. A South African ChatGPT-themed wave hit 100,000 mailboxes a day. A Claude-themed wave reached 2,000 organisations across the US, UK, and India. The brand is the lure, the payload is the same old stealer.

The FBI's 2025 internet crime report logged $20.9 billion of reported losses, a 26% rise on 2024. Investment fraud is the largest category at $8.6 billion. Business email compromise is the largest enterprise threat at $3 billion. Ransomware, by reported loss, is smaller than either. The shape of the numbers is the story.

Barclays surveyed 1,000 UK business leaders in April. Sixty-eight per cent plan to spend more on cyber security in the next year. Twenty-six per cent say AI brings new risks they cannot answer. Average spend hits £505,000, but a micro business spends £15,000 and a large one spends £1.3 million. The numbers underneath the headline are the more useful ones.

The government finally announced a statutory defence for security researchers under the Computer Misuse Act. The defence covers around 300 people. The UK has 70,000 cybersecurity professionals. The number you remember from this filing is 0.4%.

The attacker already has the password. They press the login button. Your phone buzzes. They press it again. It buzzes again. Five times. Twenty. Two hundred. At three in the morning. Eventually somebody taps approve, just to make it stop. That's how Uber lost its single-sign-on, and it's how a lot of UK firms will lose theirs.

Two-thirds of open-source maintainers do not know the Cyber Resilience Act exists. Most UK firms shipping software into the EU haven't checked whether it applies to them. The deadline is December 2027 and the obligations include something most SMEs do not yet produce: a software bill of materials.

Vulnerability exploitation is now the leading cause of breaches. AI is finding bugs faster than anyone can triage them. Time-to-exploit has gone from 840 days to under two. CISA has rewritten the federal patching rules. If your patching process still runs off CVSS scores, it is solving last year's problem.

The NCSC published two pieces of guidance in a fortnight that an SME owner can actually use. One is about agentic AI, the kind that takes actions on your behalf. The other is about zero trust network access. Both share the same underlying advice: the user's location stopped being a security signal a while ago.

The FBI has warned that the Silent Ransom Group is now sending people into law firm offices, claiming to be IT, then plugging a USB drive into a partner's machine. The script is short, reception lets them through, and the data leaves the building. The fix is mostly process, not technology.

From Bletchley Park on 27 May, the GCHQ director said the UK has a narrowing window to keep its technological edge, and announced a blueprint for an AI-driven national cyber defence. Read between the speech lines: the supply chain into critical infrastructure is the lever they actually have.

A single GitHub employee installed a trojanised Nx Console extension and around 3,800 internal repositories walked out. The interesting question isn't what GitHub will do next. It's what your editor and browser extensions can already reach.

Your inbox fills up with 2,000 newsletter confirmations in an hour. None of them are malicious. That's the point. The attacker is using the noise to hide a password reset, a fraudulent purchase, or a fake IT support call that lands moments later. A new EPFL paper has the data.

Most AI policies are vendor templates with the company name swapped in. They ban the obvious, permit the vague, and tell you nothing about how the business actually wants AI used. A coherent policy is a short one that takes a position.

The 1990 Computer Misuse Act predates the public web. Reform has been promised for six years. The May 2026 King's Speech finally put it in a bill, bundled into the National Security Bill. Here's what's likely to change and what's still vague.

A lot of guidance is telling UK businesses they need an electronic complaint form by 19 June 2026. The statute doesn't say that. It says facilitate, and gives a form as one example. Here's what's actually required and what isn't.

A year on from the April 2025 retail attacks, the numbers are in. M&S has posted £101.6 million in direct costs and a 16.4% fall in fashion sales. The Cyber Monitoring Centre put the combined bill at £270 million to £440 million. The useful lessons for an SME are the unglamorous ones.

The NCSC has published guidance on decommissioning assets. The headline is simple: things you no longer use stop being assets and start being liabilities. The boring work of switching them off is one of the highest-value security jobs most businesses skip.

Cifas surveyed 2,000 UK employees at large companies. Nearly a quarter know someone who has fiddled expenses. One in eight know someone who has sold a login. Insider risk is a culture problem before it is a tooling problem.

The NCSC has told UK organisations to prepare for a wave of urgent patches as AI accelerates vulnerability discovery. The same week, NHS England decided the answer was to make its open source repositories private. Only one of those approaches actually fixes anything.

CVE-2026-31431 lets any local user become root on Ubuntu, RHEL, Debian, SUSE, Amazon Linux, and most other distros. The exploit fits in 732 bytes of Python. The bug has been there since 2017.

CVE-2026-41940 lets an unauthenticated attacker take root on a cPanel or WHM server. It was being exploited for around a month before the patch landed. If your website lives on shared hosting, this affects you.

The 2026 UK Cyber Security Breaches Survey says 43% of businesses had an incident in the past year. Phishing was involved in 85% of those. AI hasn't changed what works, it's just lowered the price of doing it at scale.

Two-step verification on. Domain ownership protection on. GoDaddy transferred a non-profit's 27-year-old domain to a stranger in four minutes. The lesson is about the registrar layer most businesses never think about.

Two thirds of UK organisations cannot account for what staff share with AI tools. Now agentic AI is being deployed faster than anyone can govern it. The two problems are the same problem.

The NCSC has flipped its authentication advice at CYBERUK 2026. Passkeys are now the recommended default, and password plus two-step verification is the fallback. The reasoning is worth understanding.

Tickets closed. Rules written. Logs ingested. The NCSC's Dave Chismon argues most security operations metrics actively make detection worse. The one that counts is whether you spot attacks in time.

The UK's data protection regulator is being restructured under the Data (Use and Access) Act 2025. New board, new CEO, new statutory objectives. The name is the least interesting part.

The biggest overhaul of UK security regulation since 2018 is in committee. MSPs are in scope, incident reporting gets a 24-hour clock, and fines go up to £17 million. Here's what it means in practice.

The NPSA gives away a complete, professionally designed security awareness campaign kit. Posters, booklets, checklists, and a full starter guide. Most organisations don't know it exists.

CVE-2026-2441 is actively being exploited in the wild. A use-after-free bug in CSS handling means a crafted webpage is all it takes. Push the update now.

Schneier and co have reframed prompt injection as 'promptware': a full 7-stage kill chain. The uncomfortable truth: LLMs can't distinguish instructions from data. This isn't a bug you can patch.

Containment over correctness, reversibility over impact, protecting state before touching services. What your first five minutes should actually look like.

Notepad++ had its update service hijacked by state-sponsored attackers. Windows Notepad got a CVSS 8.8 command injection. Two editors, two attack vectors, same lesson.

Global Telnet scanning dropped overnight in January 2026. Days later, a critical telnetd authentication bypass was disclosed. The protocol is old. The lesson is current.

A plain-English walkthrough of the five Cyber Essentials controls, what the assessment looks like, and what it does and doesn't prove about your security.