The Security Situation of the Last Three Months – From the PHP Hole to the Microsoft Blackout

This article walks through the quarter in order. It explains what the PHP flaws are and why they matter to anyone with a website, covers the biggest incidents that followed, unpacks the Microsoft outage, and then steps back to ask what the whole period tells us. It is written for readers who are not security specialists, so each technical term is explained when it first appears.

Between late August and late November 2026, the security world moved from a steady drumbeat of breaches to a stretch that some practitioners started calling “Black October”. Four major incidents landed within weeks of each other: the ASOS app hack on 6 October, the French supermarket refrigeration attacks on 11 and 12 October, the PHP vulnerabilities made public on 21 October, and the Binance theft on 29 October. Then, in mid November, Microsoft’s cloud suite went dark for roughly a day, which was not an attack at all, yet it hurt almost as much as one.

This article walks through the quarter in order. It explains what the PHP flaws are and why they matter to anyone with a website, covers the biggest incidents that followed, unpacks the Microsoft outage, and then steps back to ask what the whole period tells us. It is written for readers who are not security specialists, so each technical term is explained when it first appears. One caution applies throughout: some attributions and loss figures are still being investigated and may be revised.

The quarter at a glance

A short timeline helps before the details. All dates are in 2026.

  • Early September: a wave of data breaches and ransomware cases, including a large US utility and a health data provider
  • 14 September: CenterPoint Energy files a formal notice about stolen customer data
  • 6 and 7 October: a US government warning about attacks on firewalls, and ransomware at a Japanese cloud provider
  • 11 and 12 October: remote sabotage of refrigeration systems in French supermarkets
  • 15 and 16 October: abuse of a backup vendor’s credential against hundreds of Microsoft 365 tenants
  • 21 October: PHP releases emergency fixes for two critical flaws
  • 26 October: US authorities confirm the PHP flaws are being exploited
  • 29 October: roughly US$6.3 billion is stolen from Binance
  • 12 to 14 November: a global Microsoft 365 outage
  • 19 November: a second large PHP security release, fixing 18 further issues

September: the quiet that was not quiet

September did not produce one defining event, but it set the tone. One review of the month noted that the incidents did not follow a neat pattern, and that they spread across critical infrastructure, law enforcement, healthcare and public services.

The utility that learned about its breach from a forum post

CenterPoint Energy supplies electricity and gas to millions of customers. It said in a filing on 14 September that it became aware of an online post by a third party claiming to hold a dataset of customer information. A forum user had claimed on 12 September to have extracted roughly 7.49 million files through an unsecured interface, a claim that the company had not independently verified. The company later said the exposure reached the personal information of more than one million people in Texas. The lesson is uncomfortable: the first public warning came from the criminal, not from the company’s own monitoring.

Health data, governments and criminals posing as IT staff

In the same month, Aesto Health said a breach affected over 9.5 million patients, and the Berlin state government dealt with stolen data being published by a ransomware group. Meanwhile, a campaign described as fake IT calls targeted executives and staff at organisations using Microsoft 365, aiming at data theft and extortion. That last item matters because it shows attackers choosing the human at the keyboard over a technical hole, a pattern that returns later in this story.

The PHP hole

The flaw at the centre of this quarter is known as PHP2Shell. To see why it caused such alarm, it helps to start with what PHP is.

Why a flaw in PHP reaches so far

PHP is a programming language that runs on web servers. Popular website platforms such as WordPress, Drupal and Joomla are built on it. Wiz Research, the firm that found the problem, describes PHP as powering about 70 percent of all websites. That figure is the firm’s own estimate, but it explains the stakes: a weakness in PHP itself is not a flaw in one product, it is a flaw underneath thousands of products at once. For anyone in Ghana running a business site, a church or school website, or a small online shop on WordPress, this is directly relevant.

What the flaws actually are

Two flaws share the label. They are tracked as CVE-2026-109618 and CVE-2026-109619, and both are critical remote code execution flaws that an unauthenticated attacker can reach. Remote code execution means a stranger on the internet can make your server run commands of their choosing. Unauthenticated means no password or account is needed first. Both score 9.2 on the CVSS v4.0 scale, where 10 is the maximum.

In plain terms, the first flaw happens when PHP runs out of memory and leaves invalid data behind for its cleanup routine, and the second lets a PHP worker reuse stale information from a previous request. Think of a restaurant kitchen where a cook who is overwhelmed leaves a dirty tool on the counter, and the next order is prepared with it. The result is chaos the attacker can steer. Researchers at Datadog reproduced the first flaw with a single unauthenticated web request against PHP 8.5.11.

From discovery to attack in about nine weeks

The timeline is as important as the bug. Wiz says it found the first flaw on 19 August, using AI models alongside its own tools, confirmed attacks against Drupal and WordPress within two days, and reported the issues to the PHP security team on 24 August. The fix was planned for later, but details leaked ahead of the schedule, the PHP team published an advance notice on 20 October, and fixed versions came out on 21 October.

Attackers moved faster than the patch. Amazon reported exploitation attempts as early as 17 October, four days before the fixes, and CISA, the US cyber agency, added both flaws to its list of known exploited vulnerabilities on 26 October. Some reporting suggested exploitation was slower than expected, yet millions of servers remained at risk. Both can be true: a weapon can be loaded and available without every attacker using it on day one.

Which servers are exposed and which versions fix it

The affected setups include Apache with mod_php, PHP-FPM, php-cgi, FrankenPHP and LiteSpeed, and PHP considers mod_php the highest risk. The fixed releases are PHP 8.5.12, 8.4.27, 8.3.36 and 8.2.35. Older branches are a separate worry. One European agency lists PHP 8.1 and older as affected, yet the project provided only minimal source level backports for the oldest versions. If you run an old site on an unsupported PHP version, upgrading the version itself is the real fix, not a patch.

The patch that was not the end

Many people assume that one update closes a story. It did not here. PHP’s 19 November release fixed 18 CVEs. An analysis of the follow up issues said that some cases were not covered by the October fix, and one more problem became reachable through a change introduced in 8.5.12. One of the November problems, CVE-2026-129289, affects all PHP branches in certain configurations running behind Nginx or Apache and can let an unauthenticated attacker execute PHP code. The risk depends on how session storage and the web server are set up, not just the PHP version number.

The practical meaning is that patching PHP is now a repeat task, not a one off event. When a core component receives intense scrutiny, more flaws tend to surface in the weeks that follow.

The other October incidents

PHP was one of several stories. Many of the others share a theme: attackers gained access through something the victim already trusted.

Binance and the US$6.3 billion theft

Attackers began draining Binance hot wallets across several blockchains at 20:05 UTC on 29 October, moved most of the funds within about 20 minutes, and Binance puts the loss at about US$6.3 billion, the largest cryptocurrency theft publicly reported. A hot wallet is a wallet connected to the internet so that customers can withdraw quickly, which also makes it a target.

The route in was long and patient. Binance’s incident report traces it to a months long campaign against its engineers, including a tampered installer in June, a fake job interview that planted a malicious code repository in September, and abuse of an internal AI tool on 20 October that led to the signing service used in the theft. In other words, the attackers did not break a vault door. They spent months befriending the people who held the keys.

The FBI attributed the theft to North Korea on 31 October and tied the laundering to a cluster called TraderTraitor. Binance says about US$3 billion has been frozen on chain and that it is covering the remaining losses itself. Binance also says it found no evidence that wallet private keys were stolen. Final totals may change, since different analytics firms use different pricing and counting methods.

ConnectWise and the credential that opened 300 doors

ConnectWise disclosed on 16 October that an attacker abused the Microsoft 365 application credential behind its Cloud Backup service, which held standing permission to read the tenants of service providers and their customers. Sign in attempts appeared in almost 1,000 tenants, with successful sign ins in more than 300. A tenant is simply one organisation’s private space inside Microsoft 365.

Microsoft assessed that a China state aligned group, Storm-0940, was responsible, though no government has publicly attributed the compromise. This is the textbook supply chain attack: instead of attacking hundreds of companies, attack the one vendor they all trust.

French supermarkets and cyberattacks that cause physical damage

Not every attack steals data. Overnight on 11 to 12 October, attackers forced remote refrigeration controllers at French supermarkets into defrost cycles. French authorities later counted about 700 affected retail sites and at least 15 fires, estimated food losses in the tens of millions of euros, and attributed the attack to Shahid Kaveh, a cyber unit of Iran’s Revolutionary Guard Corps. The attackers reached equipment that was exposed to the internet, a reminder that fridges, heating controllers and other industrial devices need the same care as laptops.

Cloud providers, firewalls and giant data exposures

The same month brought a string of other large events:

  • Ransomware hit Japan’s IDCF Cloud on 7 October, disrupting 495 companies and local governments by encrypting customer virtual machines.
  • The FBI and Secret Service warned on 6 October that the FortiBleed campaign against Fortinet firewalls was still active, with 86,644 working device credentials verified across 194 countries by SOCRadar.
  • Denmark disclosed that misuse of a private company’s lookup access exposed records on about 8.8 million people from its national register.
  • Florida’s education department said about 4.9 million people may be affected by unauthorised access to its student aid website.
  • Capcom confirmed that 11,774,247 records from its membership service were exposed.

Criminals using AI to attack

One detail from South Korea deserves attention. Police said a former penetration tester allegedly led a group that breached more than 90 organisations and stole about 8.2 million records, pairing a coding agent with an autonomous AI penetration testing system to breach about 60 of them. Attacks of this kind used to need a skilled team. AI tools are lowering that barrier.

The Microsoft blackout

Then, on 12 November, the story changed shape. The question stopped being “who attacked us?” and became “what happens when a single provider simply stops?”

What users experienced

Problems began at about 16:00 GMT on Thursday, with users unable to reach Outlook, and within about an hour the outage tracker Downdetector had reports from 74 countries. Teams and Copilot were also disrupted. It affected businesses, institutions and individuals worldwide, and it was one of the longest in Microsoft’s history. In Washington, House staff told Politico that it ended a day of legislative work.

Recovery was slow and messy. More than 12 hours in, Microsoft reported only marginal success, and servers that had been restored were failing again. Microsoft marked the incident as restored late on 13 November and put the impact window from about 11:20 AM Eastern on 12 November to 11:10 AM Eastern on 13 November. Full normality, including mail queues, followed into 14 November.

What actually caused it

Microsoft said early on that it was not the result of a cyberattack. The later explanation was a chain of small events. A Microsoft executive said three factors combined: a latent defect that made gateway servers overwhelmed and unresponsive under certain loads, and a failover of Teams traffic about fifteen hours earlier to the primary gateway that also supports Exchange Online. A routine switch of Teams traffic then increased the load on servers already running at reduced capacity, so the system in effect overloaded itself, and the barriers that should have contained the failure within separate shards did not hold.

Here is an analogy. Imagine a building where a corridor door is supposed to stop a fire from spreading. A small spark appears, but the door is jammed open, so the spark reaches the whole floor. The spark was minor. The failed containment made it a disaster.

Why an outage counts as a security story

Security professionals use a model with three goals: confidentiality (secrets stay secret), integrity (data is not tampered with) and availability (systems work when needed). Most headlines focus on the first two. This outage was purely about availability. Yet for a hospital, an airline desk or a small business that runs on email, being locked out costs the same hours and money as being attacked. It also arrived after a nearly ten hour Microsoft 365 disruption in January, adding to a troubled year for the company. One more point: during any large outage, scammers rush to send fake “service restored” or “reset your password” messages, so staff should be wary of unexpected emails when systems are unstable.

What the pattern tells us

Looking across all these events, several threads connect them, even though the incidents do not appear to be connected and no common perpetrator has been identified.

Speed is the new weapon

The PHP flaw moved from discovery to mass exploitation attempts in under two months, and attackers began probing days before the patch. The time defenders have between “a fix exists” and “attackers are using it” keeps shrinking. For site owners, that means updates should be planned and routine, not postponed to a quiet weekend.

AI is accelerating both sides

AI helped researchers find the PHP flaws, helped a criminal breach dozens of organisations in South Korea, and was part of the path into Binance’s systems. The technology does not create new kinds of weakness so much as it speeds up the search for existing ones. That favours whoever acts first, and defenders usually have to react.

Trust relationships are the weakest link

Binance fell through engineers’ laptops and fake recruiters. ConnectWise fell through one backup credential with broad permissions. Denmark’s register fell through a supplier’s lookup account. In each case the organisation’s own defences were not the point of entry. A trusted partner or a trusted person was. Broad, permanent permissions given to vendors are a quiet risk.

Identity is the new perimeter

Stolen passwords, leaked credentials and fake IT calls feature across the quarter, from FortiBleed to the Danish register claim to the Microsoft 365 social engineering. Strong multi factor authentication and tight control over who holds which permissions matter as much as any firewall.

Cyber incidents are leaving the screen

Fridges defrosting and fires starting in supermarkets are a reminder that the line between digital and physical is thin. Water systems were also targeted earlier in the year, with more than 30 Minnesota community systems affected in July. Anything connected to the internet, from cold rooms to building controls, is part of the attack surface.

Concentration risk is a security risk

Microsoft 365, PHP and a handful of cloud and backup vendors support huge numbers of organisations. When one of them has a bad day, whether through a flaw or an outage, the damage is shared widely. That does not mean avoiding them, but it does mean having a plan for the day they fail.

What to do now

Practical steps depend on who you are, so here are focused suggestions.

If you own or manage a website

Check which PHP version your hosting uses and update to a fixed release: 8.5.12, 8.4.27, 8.3.36 or 8.2.35 at minimum for the October flaws, and the November release for the newer ones. Ask your hosting provider to confirm in writing. Update WordPress, Drupal, Joomla and their plugins, remove unused plugins, and keep backups stored away from the server. If you are still on PHP 8.1 or older, plan a move to a supported version.

If you run a small business

Turn on multi factor authentication for email and finance accounts. Keep an offline or alternative way to reach staff and customers, such as a second messaging channel, in case your main email goes down. Train staff to treat unexpected calls from “IT support” with suspicion, and to verify through a known number before sharing any code or password.

If you manage IT in a larger organisation

Audit which vendors hold standing access to your systems and reduce it to what is needed. Remove internet exposure from industrial and building controllers where possible. Rehearse an outage of your main productivity suite, including how you would communicate and which tasks could continue offline. Watch your external footprint for exposed old software.

If you are an ordinary user

Use a password manager and unique passwords, switch on multi factor authentication, and be cautious after any breach notice, since stolen details are often used for convincing phishing messages. Be wary of messages claiming to fix a recent outage or breach, and go directly to the official website instead of clicking links.

Final thoughts

The last three months showed security from three angles. The PHP flaws showed how a single weakness in a widely used building block can ripple across a large share of the web, and how patching can take several rounds. The October incidents showed attackers patiently exploiting trust, in people, vendors and connected devices, with AI increasingly in the mix. The Microsoft outage showed that availability is part of security and that a small defect can cascade when containment fails.

None of this calls for panic. It calls for habits: patch quickly and repeatedly, limit what any one account or vendor can do, protect identities, and prepare for the day a service you rely on is not there. Details of several incidents are still developing, so check official advisories for the latest guidance.

Gabby
Gabby

Inspiring readers to expound the possibilities of the unfolding World

Conversation

  • Share in the conversation!
No password needed - you'll confirm your email when you post.