When news of PHP2Shell broke, some organizations responded in hours. Others took weeks. What is the difference between the most resilient companies and the least?
AI is increasing vulnerability discovery; within days of PHP2Shell, another critical vulnerability discovered by AI sent responders rushing yet again. In this environment, understanding what factors predict the likelihood of being able to quickly and efficiently respond is critical for the future. For our own organizations, this information can help us craft successful programs, and for third-parties, to model the risk in our supply chain.
To answer those questions, we studied the responses of vendors monitored in the UpGuard platform after the release of PHP2Shell. This scope is not a count of the total impact, but a focused view on companies that are part of real supply chains, where PHP2Shell was a real risk.
The results affirm that best practices for maintaining software continue to pay dividends. Organizations that were already on a maintained version of PHP were the most likely to update to a secure version. Because PHP has been around for well over 20 years, there are many older versions past end of life. End-of-life versions may still have unofficial security updates produced by those in the community, or have workarounds that can be applied to the code, but they are habitually insecure and do not have access to standard security measures.
Using maintained software is correlated with the continued ability to update that software, and vice versa.
Group by
Colour fixed by
- No visible fix
- PHP 8.2
- PHP 8.3
- PHP 8.4
- PHP 8.5+
- Fixed, later vulnerable again
- No visible fix
- Days 1–2
- 3–4
- 5–6
- 7–8
- 9–10
- 11–12
- 13
Each dot is ten companies' main websites that were running a vulnerable PHP version when PHP2SHELL was disclosed
Show the current grouping as a table
Source: UpGuard scan data. Counts are aggregated; no company is identified.
Industry, geography, size
PHP is used all over the world, across such a wide breadth of hosts as to potentially affect every type of organization. While the statistic that 70% of the world’s hosts use PHP is useful for communicating the severity of PHP2Shell to a wide audience, it belies the crucial differences in criticality of those hosts. For example, part of PHP’s prevalence is that it is part of WordPress, which is cited as accounting for 40% of the world’s websites. Those sites are overwhelmingly personal and small businesses. Modeling the risk of a PHP vulnerability requires accounting for whether it has demonstrated risk for a real supply chain.
As much as PHP2Shell gripped the collective imagination with the specter of AI vulnerabilities run amok, the data on remediation activity confirms that best practices are called that for a reason. If organizations updated to a secure version, they were likely to do so quickly; half of all fixes came within the first 48 hours and nine in ten within a week, with a slower trickle after.
Who the organization was made little difference. Grouped by sector, company size or country, the response curves sit close together. By 4 November the best and worst sectors were 10 percentage points apart, company size bands 12 points and the eight largest countries 20 points.
Share of each group's sites showing a fixed version, by days since detection. Faint lines are the other groups
Sector
4 Nov range: 42–52% (10 pts)
- Highest: Financials 52%
- Lowest: Consumer Staples 42%
Company size
4 Nov range: 41–53% (12 pts)
- Highest: 51–250 53%
- Lowest: 1–10 41%
Country (top 8)
4 Nov range: 40–60% (20 pts)
- Highest: Australia 60%
- Lowest: New Zealand 40%
| Group | Sites | Share showing a fixed version |
|---|---|---|
| Sector: Industrials | 5,588 | 47% |
| Sector: Information Technology | 2,607 | 45% |
| Sector: Consumer Discretionary | 2,454 | 46% |
| Sector: Financials | 937 | 52% |
| Sector: Health Care | 935 | 50% |
| Sector: Materials | 580 | 48% |
| Sector: Consumer Staples | 290 | 42% |
| Sector: Telecommunication Services | 227 | 43% |
| Sector: Utilities | 158 | 49% |
| Sector: Energy | 100 | 49% |
| Company size: 1–10 | 3,724 | 41% |
| Company size: 11–50 | 3,646 | 49% |
| Company size: 51–250 | 2,455 | 53% |
| Company size: 251–1K | 869 | 50% |
| Company size: 1K–5K | 253 | 49% |
| Country (top 8): United States | 4,831 | 46% |
| Country (top 8): Australia | 1,958 | 60% |
| Country (top 8): India | 680 | 51% |
| Country (top 8): Canada | 609 | 45% |
| Country (top 8): United Kingdom | 568 | 54% |
| Country (top 8): Brazil | 432 | 46% |
| Country (top 8): New Zealand | 305 | 40% |
| Country (top 8): South Africa | 278 | 55% |
One line per group of 100+ sites; company size excludes sites with no size data. Source: UpGuard scan data.
Better is better
The distinction holds not just between maintained and EOL versions, but across the continuous distribution of how old a starting release version was. Among sites on a maintained version, those already on the latest release before the fix had fixed 96% of the time by 4 November, compared with 87% for sites two or three releases behind, 61% for four to ten behind and 48% for eleven or more behind.
Among end-of-life sites, the chance of moving to a maintained version fell the longer the version had been out of support, from 19% on PHP 8.1 (under a year past end of life) to 3% on PHP 5.x and older (about eight years). Where end-of-life sites did migrate, they were only somewhat slower than sites patching a maintained version: a median of 2.6 days after detection, compared with 1.7 to 2.3 days.
If this seems obvious, it is: maintaining software makes it easier to apply the next update when a critical vulnerability requires it. But it should also make tangible the cost of that debt. Being behind is not an abstract gap; it is a measurably longer exposure window when the next vulnerability hits.
Maintained versions, by releases behind at detection
Share showing a fixed version, by days since detection
- On the latest release · 4,726
- 2–3 releases behind · 817
- 4–10 releases behind · 368
- 11+ releases behind · 396
End-of-life versions, migrated by 4 Nov, by years past end of life
Share migrated, with the 95% confidence interval
| Group | Sites | Share | 95% confidence interval |
|---|---|---|---|
| Maintained versions, On the latest release | 4,726 | 96% showing a fixed version | n/a |
| Maintained versions, 2–3 releases behind | 817 | 87% showing a fixed version | n/a |
| Maintained versions, 4–10 releases behind | 368 | 61% showing a fixed version | n/a |
| Maintained versions, 11+ releases behind | 396 | 48% showing a fixed version | n/a |
| End-of-life PHP 8.1 | 1,530 | 19% migrated (290 sites) | 17%–21% |
| End-of-life PHP 8.0 | 601 | 16% migrated (97 sites) | 13%–19% |
| End-of-life PHP 7.4 | 3,246 | 11% migrated (347 sites) | 10%–12% |
| End-of-life PHP 7.0–7.3 | 1,208 | 6% migrated (67 sites) | 4%–7% |
| End-of-life PHP 5.x and older | 984 | 3% migrated (33 sites) | 2%–5% |
Releases behind counts from the fixed release of each maintained branch (8.2.35, 8.3.36, 8.4.27, 8.5.12). Years past end of life is the median for each group of branches; shading is the 95% confidence interval. Source: UpGuard scan data.
Who is still at risk
As a result, what remains exposed is almost entirely end-of-life PHP. End-of-life sites were 55% of affected sites at detection and 91% of those still showing a vulnerable version two weeks later. Some of them may have an unofficial patch that leaves the version number unchanged, which scan data cannot see.
One challenge that these host maintainers face is that they may be applying mitigations that cannot be observed externally. In the case of PHP2Shell, security scanners like UpGuard cannot safely test for the vulnerability directly, and thus must rely on advertised software versions. While imprecise in the detection of PHP2Shell exploitability, the weight of evidence indicates that these organizations should find a way to move to a maintained branch. PHP2Shell just provides one more reason to follow what were already good practices.
End-of-life share of the sites still showing a vulnerable version, by days since detection
| Point | End-of-life share | End-of-life sites | Maintained-version sites |
|---|---|---|---|
| Detection | 55% | 7,569 | 6,307 |
| 4 November | 91% | 6,737 | 643 |
Same sites as the first chart. Source: UpGuard scan data.