Most traffic reaching a public website today does not come from a person. Thales counted bots as 53% of the web traffic it observed in 2025, and bad bots as 40%. Much of that automation walks the internet looking for something it can use: an outdated extension, a login page with no limits, a configuration file in the wrong folder. This guide explains how that scanning leads to defacement, data loss and downtime for government agencies and large groups, why unprotected sites are found first, and what to put in place.
Most of your visitors are machines
The 2026 Thales Bad Bot Report, based on full-year 2025 activity observed by Thales, puts automated traffic at 53% of the total and human traffic at 47%. Bad bots rose from 37% to 40% in a year. Thales reports that daily AI-driven bot attacks grew from about 2 million to 25 million between 2024 and 2025, and that 41% of bot attacks used the Chrome browser to look like legitimate traffic.
These are one vendor's figures, not a census of the internet, but the direction is consistent. Verizon's 2026 Data Breach Investigations Report describes traffic from AI crawlers growing 21% month on month while human-led traffic stayed flat.
The access logs of any public website show what this means: requests all day for administration pages the site does not have, for files such as .env or backup.zip, and for upload endpoints of extensions that were never installed. Nobody chose your organisation. The tools ask every address the same questions and keep a list of whoever answers.
What automated scanners look for
The checks are cheap to run and the list is short:
- Exposed administration and login pages, tried with default, leaked or guessed passwords.
- Outdated content management systems, extensions and themes with published vulnerabilities, identified from version strings and file paths.
- Leaked configuration and backup files: environment files, database dumps, archives and source control folders reachable from the web.
- Credential stuffing targets: any login form that accepts unlimited attempts, fed with passwords stolen in other breaches.
- Forgotten hosts: staging sites, old campaign microsites and test subdomains nobody patches.
- Unprotected APIs. Thales found 27% of bot attacks in 2025 were aimed at API endpoints.
For an agency or a group with dozens of brands and regional sites, the weakest host sets the exposure for all of them. Hacktivists may choose government targets for political reasons, but discovery is automated and indiscriminate.
How fast a new vulnerability becomes an attack
The gap between a flaw becoming public and attackers using it is now measured in days, and often it is negative.
VulnCheck identified 884 vulnerabilities first exploited in 2025. For 28.96% of them, exploitation was seen on or before the day the CVE was published. Network edge devices were the most targeted technology, followed by content management systems, which are commonly exposed to the internet.
Mandiant's M-Trends 2026 found exploits were the most common way in for the sixth year running, at 32% of the intrusions it investigated in 2025. It estimates the mean time to exploit at minus seven days: on average, exploitation began before a patch existed. Flaws exploited before any fix pull that average below zero, so not every flaw is attacked early, but you cannot plan around a comfortable gap.
Defenders are slower. In the 2026 DBIR, exploitation of vulnerabilities became the most common initial access vector for breaches, at 31%, overtaking credential abuse for the first time in 19 years. Only 26% of vulnerabilities in CISA's Known Exploited Vulnerabilities catalogue were fully remediated in 2025, and the median time to resolution rose to 43 days. Exploitation arrives in days; patching a large estate takes weeks. Those weeks are when public websites get defaced.
From a scan to a defaced homepage: the 2026 CMS campaign
On 10 July 2026, Belgium's Centre for Cybersecurity (CCB) warned of a worldwide mass defacement campaign against public-facing websites running outdated content management software. Attackers dropped files into web roots, template folders and image directories. The CCB described the targeting as broad and opportunistic: any internet-facing site running a vulnerable extension was at risk, whatever its sector.
The advisory links the campaign to three extension vulnerabilities rated CVSS 10.0. Each lets an unauthenticated attacker upload and run code on the server in a single scripted HTTP request. For the first, the fix shipped on 3 June 2026, working exploit code was public from 9 June, and CISA listed it as known exploited on 16 June. A second had been exploited as a zero-day, sometimes followed by a rogue super user account. CISA added it and the third to the catalogue on 7 July.
The CCB's most important point concerns what follows. The access needed to change a homepage is usually enough to plant a hidden web shell or administrator account, take data, move into connected systems or inject phishing pages. Patching protects against future exploitation but "does not remediate historic compromise." CISA's Binding Operational Directive 26-04, referenced in its 7 July alert, likewise sets expectations for when US federal agencies must check whether a system was compromised before the patch was applied.
How bots cause downtime, with or without a breach
A site does not have to be breached to go offline. Scanning, credential stuffing and scraping hit the pages that are most expensive to serve: login, search, forms and anything that queries a database. Enough of them and the origin slows or stops, and real users are locked out.
Deliberate denial of service sits on top of that background load. In its January 2026 alert, the UK's National Cyber Security Centre said Russia-aligned hacktivist groups continue to try to take websites offline and disable services, including frequent DDoS attempts against UK local government. The ENISA Threat Landscape 2025 found public administration to be the most targeted sector in the EU incidents it analysed.
For companies, the cost has been measured. A 2024 Splunk and Oxford Economics study of Global 2000 companies put unplanned downtime at $400 billion a year, or 9% of profits, with 56% of incidents caused by security events. For an agency the cost is citizens who cannot file, pay or find urgent information, and lost trust that takes longer to repair than the server.
Why unprotected sites are found and hit first
Automated attacks run on economics. A tool that sends the same few hundred requests to millions of hosts does best against sites that answer everything quickly, at any volume, straight from the application: a login form with no attempt limit, an origin that renders every page on demand, an application that receives exploit payloads unfiltered.
A site that throttles repeated requests, blocks known exploit patterns and challenges suspicious automation is slower, noisier and less rewarding to attack. That does not make it immune. It means an opportunistic campaign gets less from it than from the next address on its list, and a determined attacker has to work harder and is more likely to be seen.
What edge protection does
Edge protection, usually delivered on a CDN, inspects and filters traffic before it reaches your servers. BrandPort's Firewall and traffic limits and Website and government firewall services work at this layer. The controls that matter most are below.
Rate limiting
Rate limiting caps how many requests a client can make to a given path in a given time. Apply it first where abuse is cheap for an attacker and costly for you: login and password reset, search, contact and application forms, and API endpoints. Tighter limits on sensitive paths, combined with a challenge before outright blocking, slow credential stuffing to a pace that is easier to spot and far less productive, and keep scrapers from exhausting the origin. Thales warns that IP reputation and rate limits alone are not enough, because bad bots now use real browsers and residential proxies, so treat rate limiting as one layer.
Web application firewall rules
Managed WAF rules block common attack classes such as injection and path traversal. Their most valuable use during a campaign is virtual patching: when a flaw is disclosed, a rule that blocks requests to the vulnerable endpoint protects every site behind it while teams test and deploy the real fix. The CCB recommends exactly this, a WAF that filters anomalous requests to extension endpoints. Edge rules can also refuse requests for files that should never be public, such as environment files, archives and source control folders, and keep administration paths reachable only from approved networks or through zero-trust access.
Bot filtering
Bot management separates verified good bots, such as the search crawlers you want, from automation you do not. Hosting-network traffic rarely has reason to submit a citizen form or log in to a staff portal, so challenge or close those paths to it, and show human verification only when behaviour looks automated.
A CDN and caching to stay online
A CDN serves cached copies of public pages from locations close to each visitor, so a surge in traffic, hostile or not, is absorbed at the edge instead of at your origin. The NCSC recommends considering a CDN for web-based services as an upstream denial of service defence. Lock the origin to accept traffic only from the CDN so attackers cannot bypass the edge, and prepare a static fallback page the edge can serve if the application has to come offline.
What the edge cannot do
Edge controls do not fix vulnerable code. They cannot tell a stolen but valid password from the real user without other signals, they do not see what an attacker does with an administrator account, and no rule exists for a flaw nobody has disclosed. That is why the practices below still decide the outcome.
Practices that still decide the outcome
- Patch by exploitation, not only by score. Track known exploited vulnerability catalogues and treat a listed flaw on an internet-facing system as an emergency. Remove extensions and themes you do not use.
- Assume compromise before the patch. After fixing an exploited flaw, look for web shells, unexpected files in upload directories and new administrator or editor accounts.
- Use multi-factor authentication on every publishing account, phishing-resistant for privileged users, with few administrators and least privilege for everyone else.
- Take administration off the open internet. Restrict it by network or put it behind zero-trust access.
- Block script execution in upload and media directories at the web server level, as the CCB advises.
- Monitor content and files with scheduled checks on key pages and integrity monitoring on the web root, alerting someone who will act out of hours.
- Keep isolated, automated backups outside the primary hosting environment and test restores. Mandiant reports that ransomware groups now go after backups and delete backup objects from cloud storage.
- Rehearse the incident playbook: preserve evidence, isolate the system, serve the fallback page, remove web shells and rogue accounts, rotate credentials, rebuild from a clean backup, and agree who speaks to the public.
Website protection checklist
| Control | What good looks like |
|---|---|
| Asset inventory | Every public site, subdomain and staging host listed with an owner, software versions and installed extensions. |
| Exploited-flaw patching | Known exploited vulnerabilities on internet-facing systems fixed or virtually patched within days, with unused extensions removed. |
| Rate limits | Limits on login, password reset, search, forms and APIs, with a challenge before blocking. |
| WAF rules | Managed rules on, virtual patches ready for new disclosures, sensitive file paths refused at the edge. |
| Bot filtering | Good bots verified, hosting-network traffic challenged on human-only paths, human verification for suspicious behaviour. |
| CDN and origin | Public pages cached at the edge, origin accepting traffic only from the CDN, static fallback page prepared. |
| Admin access | MFA on every publishing account, few administrators, admin paths off the open internet. |
| Upload directories | Script execution blocked in upload, media and temporary folders. |
| Monitoring | Content checks and file integrity alerts on key pages, routed to an on-call owner. |
| Backups and playbook | Isolated, tested backups and a rehearsed procedure for defacement and denial of service. |
Frequently asked questions
Why do bots target government websites?
Mostly because bots target everything. Scanners test every reachable address for known weaknesses. Government sites add a political motive for hacktivists and a trusted audience for phishing, which is why public administration features so heavily in threat reporting.
Does rate limiting stop bot attacks?
It stops the cheapest ones and slows the rest. Limits on login, search, forms and APIs blunt credential stuffing and scraping and protect the origin. Bots using real browsers and residential proxies can stay under simple limits, so combine rate limiting with WAF rules and bot filtering.
Can a web application firewall prevent website defacement?
It reduces the risk, especially through virtual patching that blocks requests to a newly disclosed vulnerable endpoint while you patch. It cannot fix vulnerable code or stop an attacker using a valid administrator password, so defacement defence also depends on patching, MFA, least privilege and file monitoring.
How do we prevent website downtime during an attack?
Serve public pages from a CDN cache so spikes are absorbed at the edge, lock the origin to accept traffic only from the CDN, rate-limit expensive paths, and prepare a static fallback page. The UK NCSC also advises planning for graceful degradation and keeping administrative access during an attack.
How do we start with BrandPort?
Request a briefing. BrandPort is offered by invitation. Tell us about your public websites and how they are protected today, and we will follow up.
Request a briefing
By invitation. If you want to review how your public websites stand up to automated scanning, bot traffic and denial of service, tell us about your estate and we will follow up.
Request a briefing