Stop Sending Sensitive PDFs to
Third-Party Servers
Run the full PQ PDF platform inside your infrastructure — scan, sanitize, convert, and process documents with zero external exposure.
PQ PDF is not a SaaS product. It’s infrastructure.
Privileged documents and discovery files cannot leave firm infrastructure. Attorney-client privilege doesn’t survive third-party processing.
PHI cannot pass through third-party processors without a BAA — and a BAA doesn’t prevent a breach. On-premise eliminates the exposure entirely.
Regulated filings, client contracts, and trading documents must stay within your systems. Cloud processors create a data processor relationship that regulators scrutinise.
Data sovereignty by statute. Air-gapped environments. No internet dependency once deployed. Documents classified at any level stay inside your perimeter.
If any of the above applies, cloud PDF tools are not an option. See how PQ PDF runs inside your environment →
Why teams deploy PQ PDF internally
Every file processed by a third-party cloud service is an exposure event. On-premise removes that exposure entirely.
Your documents never leave your network. No vendor breach can expose them. Your perimeter is the only boundary that matters.
- Zero vendor access to file content — ever
- Air-gapped deployment supported
- Audit logs under your control
- Four-layer processing sandbox — your configuration
Fixed deployment cost, not a subscription that grows with every hire. Per-seat SaaS runs $10–$30/user/month — $18,000/year for 50 people, more for every new staff member. On-premise is the same cost whether you have 50 users or 500.
- Unlimited internal users — no per-seat fee
- Fixed annual support, no surprise tiers
- One breach event can dwarf years of on-prem costs1
- No feature paywalls or scaling penalties
Your legal and compliance team owns the full data flow. No third-party data processor to contract, no cross-border transfers to justify — GDPR, HIPAA, and NIS2 obligations become straightforward.
- No third-party data processor agreement required
- Data stays in your jurisdiction — always
- GDPR Art. 83(5): fines up to €20M / 4% global turnover2
- HIPAA BAA not required — PHI never leaves your systems
Every processing engine is open-source. Ghostscript, Tesseract, PyMuPDF, ClamAV, YARA — no proprietary black box. You can audit exactly what runs on your files, and extend it for your workflows.
- Open-source stack — fully auditable, no black boxes
- Configure limits and parameters to your requirements
- Integrate with internal systems and identity providers
- No dependency on our uptime or pricing decisions
Security Architecture
The platform is built around a zero-trust file lifecycle. These are not marketing claims — they are verifiable in the technical security reference.
Every request creates one isolated temp directory (0700 permissions, 24-hex CSPRNG suffix).
Files are written only inside it. When processing completes, cleanup() is called
immediately after readfile() begins streaming — the directory is deleted
while the download is still in flight. No retention window, no cleanup cron, no object storage.
Error paths also trigger cleanup before returning.
Files pass through layered isolation before any engine touches them:
CSPRNG-randomised temp directory isolation; shell-escaped arguments via escapeshellarg();
hard 120-second OS-enforced process timeout; and network-namespace isolated dynamic sandbox
(unshare + strace) for behavioural analysis. No user-controlled string
ever reaches the shell interpreter unescaped.
TLS 1.3 only (1.0/1.1/1.2 disabled), cipher TLS_AES_256_GCM_SHA384.
HTTP/3 + QUIC v1 — encrypted at the transport layer, connection migration, 0-RTT
resumption with forward secrecy.
Post-quantum hybrid key exchange: X25519MLKEM768 (classical + ML-KEM-768)
where supported — your documents are protected in transit against future cryptographic threats, not just today’s.
HSTS max-age=63072000; includeSubDomains; preload (2 years).
Per-request CSP nonces (128-bit entropy) — no unsafe-inline, no unsafe-eval.
Full security header stack: X-Frame-Options: DENY, X-Content-Type-Options: nosniff,
Referrer-Policy, Permissions-Policy, CORP, COEP, NEL, Report-To —
A+ rating on securityheaders.com.
The PDF Forensics Scanner runs 44 independent analysis engines: structural integrity, byte-pattern matching (YARA), metadata (ExifTool), font analysis, CVE patterns, qpdf, PeePDF, dynamic behavioural sandbox (strace + Linux namespaces), ClamAV (700k+ signatures), ML anomaly detection (IsolationForest + RandomForest + LightGBM with SHAP), six-parser differential analysis, polyglot detection, JavaScript AST deobfuscation, offline threat intelligence (6.4M+ indicators), signature forensics, phishing detection, embedded file analysis, campaign attribution (TLSH + pHash), weighted correlation engine (60+ compound patterns), PDF token obfuscation detection, XFA FormCalc parser, action dependency graph, OCG layer cloaking, Unicode/invisible text, trailer chain forensics, codec exploit validation, entropy topology, image steganography, PDF/A compliance fraud, JavaScript behavioural emulation, font CharString emulation, XRef integrity graph, and a self-hosted Qwen 2.5 1.5B AI forensic report engine (llama.cpp, WireGuard-isolated). No file content is stored; only anonymous structural feature vectors are used for model training.
The Office Document Forensics Scanner covers Word, Excel, PowerPoint, Outlook, Access, and Visio files across 23 independent engines: container integrity, VBA macro extraction (olevba · mraptor · pcodedmp), Excel 4.0 XLM/DDE chain analysis, OLE compound structure inspection, IOC extraction (URLs · IPs · domains · registry keys · base64 payloads), ClamAV antivirus, YARA rule engine, offline threat intelligence (URLhaus · MalwareBazaar · ThreatFox · FeodoTracker), LibreOffice behavioural rendering, isolation chamber detonation (unshare + strace), entropy & compression anomaly detection, OOXML relationship forensics (remote template injection detection), metadata provenance, OPC rule validation, OOXML schema validation, font & theme forensics, MIME/transport forensics, digital signature forensics, NLP social engineering classifier (regex + LLM), intelligent cross-engine correlation (MITRE ATT&CK–mapped attack chain detection), and AI forensic report (Qwen 2.5, structured verdict + confidence + MITRE mapping). 4-mode sanitize: convert to PDF, strip macros, strip metadata, or convert to OOXML. Enterprise deployment removes the 10 MB file-size limit.
Upload two PDF or Office documents to compare their structural security profiles side by side. Both files are scanned in parallel through all applicable forensic engines, then diffed across 25+ security features — encryption status, ClamAV, YARA matches, threat intel hit, macro presence, auto-exec, IOC counts, risk score, sandbox behaviour, and more. Returns a similarity percentage, variant verdict (IDENTICAL / NEAR_IDENTICAL / SIMILAR / PARTIALLY_SIMILAR / DIFFERENT), and a differences-first table. Useful for detecting malware variants, comparing suspicious attachments, or verifying document integrity. Supports cross-format comparison (e.g. PDF vs Word). Enterprise deployment removes the 10 MB per-file limit.
PQC mode performs all key generation and wrapping in the browser
using the @noble/post-quantum library before the file is uploaded.
31 post-quantum algorithms are supported, including NIST-standardised ML-KEM-1024 (FIPS 203),
HQC-128/192/256, FN-DSA variants, and hybrid classical+PQC modes.
The server receives only an already-encrypted .pqcpdf bundle — the plaintext
file never traverses the network in PQC mode, and the server has no access to private keys.
NIST finalised ML-KEM (FIPS 203), ML-DSA (FIPS 204), and SLH-DSA (FIPS 205) on 13 August 2024.3
Every processing engine is open-source running locally: Ghostscript, Poppler (pdfunite, pdftoppm, pdftotext),
qpdf, LibreOffice, ImageMagick, Tesseract 5, PyMuPDF, ExifTool 12, YARA 4.5, ClamAV 1.4+,
PeePDF 0.4, strace + unshare, scikit-learn, acorn (Node.js).
No file data is forwarded to any external API. The only outbound call from processing
engines is ClamAV pulling signature updates from clamav.net — no file content transmitted.
The cloud service also exposes a REST API at api.pqpdf.com
for programmatic access; on-premise deployments remove this entirely, keeping all
processing inside your own network perimeter.
What changes when you deploy PQ PDF
Same 45 tools. Full capability. Zero external exposure.
The Cost & ROI Case
On-premise has a higher upfront cost. Here is why the total cost of ownership often favours it.
- 50 users → $6,000–$18,000/year
- 100 users → $12,000–$36,000/year
- Cost scales linearly with every hire
- Files processed on vendor servers
- Vendor breach = your document exposure
- Third-party data processor under GDPR — requires DPA
- Renewal risk: pricing changes, feature paywalls
- Unlimited internal users — no per-seat fee
- Cost does not rise when you hire
- Files never leave your network
- No third-party data processor relationship
- Vendor breach cannot affect your documents
- No DPA required for internal PDF processing
- Fixed annual support contract, no surprise tiers
Out-of-bounds write vulnerability enabling arbitrary code execution. Confirmed exploited in the wild by a North Korean APT group per Google Project Zero. Added to CISA's Known Exploited Vulnerabilities catalogue in September 2023, mandating federal agency remediation. Adobe security bulletin APSB23-34.14
Use-after-free vulnerability allowing remote code execution. Active exploitation confirmed. Added to CISA KEV in October 2023. NVD CVSS score: 7.8 (High). Affects both Windows and macOS deployments. Attackers weaponised PDFs to gain code execution on victim machines silently.14
Check Point Research (May 2024) confirmed multiple threat actor groups exploiting a logic flaw in Foxit PDF Reader to deliver malware including Agent Tesla, Remcos RAT, NanoCore, and ransomware droppers. The flaw involves misleading dialog prompts that cause users to approve malicious execution. CIS advisory issued.15
Discovered by EXPMON (December 2024): a zero-day in both Adobe Acrobat/Reader and Foxit PDF Reader allowed PDFs to silently leak Windows NTLM authentication hashes to attacker-controlled servers simply by being opened — no user interaction beyond opening the file required. Both vendors patched after responsible disclosure.16
Resecurity and BleepingComputer (January 2026) reported a sophisticated Windows malware framework distributed via weaponised PDF lures and deployed on the network of a Fortune 100 company. PDFSider uses DLL side-loading for AV/EDR evasion and was linked to multiple ransomware groups. HP Wolf Security (September 2025) separately documented ultra-realistic AI-generated PDF invoice lures capable of deceiving trained security staff. The sophistication of PDF-based attacks is accelerating with AI-generated content removing traditional quality indicators used to detect phishing.17
A zero-day in MOVEit, a third-party file transfer SaaS, gave the Cl0p ransomware group access to files at Shell, EY, PwC, Siemens Energy, the US Department of Energy, BBC, British Airways, and thousands more. No amount of internal security protected organisations that routed documents through the compromised service. Multiple class-action settlements in the millions are ongoing; Progress Software stated it “cannot reasonably estimate a range of possible losses” (SEC filing, 2024).5
Change Healthcare processes medical claims and documents for a significant share of US healthcare. ALPHV ransomware operators exfiltrated data affecting 190 million individuals — the largest healthcare data breach in US history. UnitedHealth Group reported $2.2 billion in direct response costs and $3.1 billion in total impact in SEC quarterly filings and their full-year 2024 earnings release (January 16, 2025).6 Regulatory investigations by HHS OCR and state AGs are ongoing.
Cl0p exploited a zero-day (CVE-2023-0669) in GoAnywhere MFT, a managed file transfer SaaS. Confirmed victims include Hitachi Energy, Procter & Gamble, Rubrik, Rio Tinto, City of Toronto, Saks Fifth Avenue, and the UK Pension Protection Fund. Fortra's $20M MDL settlement received final approval. Virtual mental health provider Brightline (a GoAnywhere victim) separately paid $7M — final court approval February 2025.21
Cl0p exploited multiple zero-days in Accellion's legacy File Transfer Appliance used by ~300 organisations. ~100 were breached including ASIC (Australian Securities regulator), Bombardier, Royal Dutch Shell, Kroger, and multiple universities. Accellion's $8.1M class action settlement and Kroger's $5M pharmacy records settlement are separately documented. This was the same Cl0p group that later hit GoAnywhere and MOVEit.22
Anthem's 2014 breach exposed the health records of 78.8 million people. Three separate enforcement actions: $115M federal class action (Judge Koh, N.D. Cal., final approval August 2018); $48.2M multistate AG settlement ($39.5M + $8.7M CA, announced 2020); $16M HHS OCR HIPAA settlement (record at the time, October 2018). Total documented liability: $179.2M+ excluding internal remediation costs.18
One of hundreds of downstream organisations breached via the MOVEit file transfer service. NSC's $9.95M settlement received US federal court final approval on May 13, 2025. The MDL in the District of Massachusetts consolidates 144+ class actions against Progress Software, NSC, and other downstream victims. Progress Software's own 10-K states it “cannot reasonably estimate a range of possible losses” from the consolidated litigation.5
Equifax's breach exposed the financial records of 147 million Americans. The FTC/CFPB consent order required a guaranteed minimum of $575 million — $425M consumer fund, $175M to 48 states and territories, $100M CFPB civil penalty — with a ceiling of $700M if the consumer fund was exhausted. FTC press release, July 22, 2019.7
A misconfigured AWS S3 bucket exposed the files of 106 million customers. The OCC issued an $80M civil money penalty (consent order, August 2020) for failure to establish an effective risk assessment process before migrating data to cloud. The $190M class action settlement received federal court final approval in 2022.8
A Magecart-style supply-chain attack in 2018 diverted customers to a fraudulent site, intercepting payment card and booking data. The ICO's Final Penalty Notice (October 16, 2020) imposed £20M under GDPR — the first major ICO enforcement under the regulation. The initial notice of intent was £183M; the final imposed fine was £20M after mitigating factors.9
Morgan Stanley hired unqualified vendors to decommission data centre hardware, leaving 15 million customers' PII on devices resold online — encryption was available but never activated. OCC consent order #2020-134 ($60M), SEC enforcement action #2022-168 ($35M, Regulation S-P violations), $60M federal class action settlement (2022), and $6.5M NY AG multistate settlement (2023). Four separate enforcement actions for one category of failure.10
The Irish DPC adopted its final decision on May 12, 2023, implementing EDPB Binding Decision 1/2023 (Article 65 dispute resolution). Violation: Article 46(1) GDPR — transferring EU/EEA personal data to the US without adequate safeguards post-Schrems II. Fine: €1,200,000,000. This demonstrates that data transfer violations — not just security breaches — attract maximum GDPR enforcement. Official source: dataprotection.ie; edpb.europa.eu.20
All figures above are documented from public regulatory orders, SEC filings, court records, and official press releases — not industry estimates. MOVEit and Change Healthcare are highlighted because they involve third-party file transfer and document processing services specifically. Individual outcomes vary; these are cited cases, not predictions. See references 5–10 below.
Compliance Framework Coverage
On-premise deployment eliminates the third-party processor relationship that creates compliance complexity across these frameworks.
When files are processed on-premise, there is no third-party data controller or processor to contract under Article 28. Data stays within the EU (or wherever you deploy). No cross-border transfer issues. Your DPO has a clear, auditable data flow with no external dependencies.
PHI processed on your own HIPAA-compliant infrastructure does not require a Business Associate Agreement with a third-party PDF service. You control the administrative, physical, and technical safeguards entirely. No vendor audit required for this workload.
Annex A controls for data classification, access control, and supplier relationships are simpler when document processing runs on infrastructure you already manage and audit. On-premise eliminates a supplier risk entry for document processing workflows.
For essential and important entities under NIS2, supply chain security is a mandatory risk management measure. Processing sensitive operational documents through an external SaaS increases supply chain exposure. On-premise removes this vector.
Post-Brexit UK GDPR mirrors EU GDPR obligations around data processors and transfers. On-premise deployment keeps document data within UK jurisdiction, removing any adequacy decision dependency for international transfers.
For classified, OFFICIAL-SENSITIVE, or operationally restricted environments, air-gapped deployment is supported on request. No part of the platform requires internet access for operation once installed.
Public UI vs Enterprise Deployment
The public UI stays free — always. Enterprise is a separate, paid on-premise installation.
| Feature | pqpdf.com (Public) | Enterprise (On-Premise) |
|---|---|---|
| Cost | ✔ Free — always | Paid installation & support contract |
| Account required | ✔ Never | ✔ Never (same UI) |
| File size limit | 50 MB per file 10 MB for forensic scanner |
Configurable — no hard cap |
| Rate limiting | Per-session: 10 ops / 5 min | None — full throughput |
| Data retention | ✔ Zero — files deleted during download | ✔ Zero — same behaviour |
| Data location | PQ PDF servers | Your servers — never leaves your network |
| All 45 tools | ✔ Yes | ✔ Yes |
| 44-Engine Forensics Scanner | ✔ Yes | ✔ Yes |
| Post-Quantum Encryption (31 algorithms) | ✔ Yes | ✔ Yes |
| ML anomaly detection | ✔ Yes (shared model) | Your own isolated model, trained on your scan data |
| Support & maintenance | Community / self-serve | Included in contract |
| Air-gapped deployment | Not applicable | Supported on request |
| Third-party data processor | PQ PDF (requires consideration) | None — you are the processor |
How PQ PDF Compares to Cloud PDF SaaS
A factual comparison against the category of cloud-hosted PDF processing tools. We do not fabricate features or misrepresent competitors — only verifiable differences are listed. See references below.
| Capability | Typical Cloud PDF SaaS | PQ PDF On-Premise |
|---|---|---|
| Data location | Vendor's cloud — region varies by provider | ✔ Your own servers — your jurisdiction |
| Pricing model | Per-seat subscription ($10–$30/user/month typical) | ✔ One-time install + fixed annual support |
| Post-quantum encryption | Not available in mainstream tools as of 20254 | ✔ 31 PQC algorithms including NIST ML-KEM-1024 (FIPS 203)3 |
| PDF threat analysis | Basic or not available | ✔ 20-engine scanner: YARA, ClamAV, dynamic sandbox, ML anomaly detection |
| Air-gapped deployment | Not available — requires internet | ✔ Supported on request |
| Processing engine transparency | Proprietary — engines undisclosed | ✔ Fully open-source stack, publicly listed |
| File retention policy | Varies — typically minutes to hours in cloud storage | ✔ Zero — deleted during response stream |
| Account required | Usually required for business tiers | ✔ Never — same as the public UI |
| GDPR third-party processor | Yes — DPA required with vendor | ✔ None — you process your own data |
| CSP / application security hardening | Varies — rarely disclosed | ✔ Per-request nonces, no unsafe-inline, full header suite — documented |
⚠️ “Typical cloud PDF SaaS” describes the category, not any single vendor. Specific features, pricing, and data handling practices vary between providers and change over time. Verify independently before making procurement decisions.
How It Works
Four steps from initial conversation to a running installation on your infrastructure.
Who It’s For
Any organisation that needs the tools but can’t send documents to a third-party server — or needs more capacity than the public UI provides.
What Stays Free — Always
- pqpdf.com will always be free. No accounts, no signups, no data retention — ever.
- Every tool on pqpdf.com remains available to everyone, always free, with no feature gating.
- Need to process files over 50 MB or scan beyond the 10 MB scanner limit? On-premise removes both constraints.
Deployment & Licensing Model
PQ PDF is deployed inside your infrastructure with a one-time setup and configuration fee, followed by an annual support and maintenance licence. Licensed per organisation, not per user — what scales is your deployment footprint.
Single site. Small team. One server or a primary + failover pair.
- ✓ Unlimited users
- ✓ Dev & staging instances included
- ✓ All 45 tools, no feature gates
- ✓ Installation & configuration assistance
- ✓ Standard support
- ✓ One-time setup fee + annual support licence
Mid-size organisation. Multiple environments — production, DR, staging, dev — or a small multi-site rollout.
- ✓ Unlimited users
- ✓ Up to 10 instances (any mix of prod / DR / staging)
- ✓ All 45 tools, no feature gates
- ✓ Full installation & configuration across all instances
- ✓ Priority support
- ✓ One-time setup fee + annual support licence
- ✓ Security patches & updates included
Large organisation. Global deployment, multiple sites, container orchestration, or air-gapped environments.
- ✓ Unlimited users
- ✓ Unlimited instances — any number of sites, regions, or containers
- ✓ All 45 tools, no feature gates
- ✓ Full rollout across all environments
- ✓ Dedicated support & direct escalation
- ✓ Air-gap deployment supported
- ✓ One-time setup fee + annual support licence
- ✓ Security patches, updates & new tools
Technical Validation
These are not claims about planned features. They describe what runs in production on pqpdf.com today — every engine version is verifiable, every architectural property is documented in source.
| Tool | Median (ms) | Avg (ms) | 7-day Uptime |
|---|---|---|---|
| a11y | 307 ms | 301 ms | 100.0% |
| camera-scan | 515 ms | 533 ms | 100.0% |
| color-inspect | 357 ms | 364 ms | 100.0% |
| compare | 660 ms | 669 ms | 98.7% |
| compress | 262 ms | 267 ms | 100.0% |
| convert | 105 ms | 105 ms | 100.0% |
| delete-pages | 259 ms | 263 ms | 100.0% |
| deskew | 308 ms | 315 ms | 100.0% |
| edit | 305 ms | 310 ms | 100.0% |
| esign | 134 ms | 135 ms | 100.0% |
| excel-to-pdf | 1,659 ms | 1,682 ms | 100.0% |
| extract-pages | 258 ms | 261 ms | 99.3% |
| extract-text | 107 ms | 109 ms | 100.0% |
| file-compare | — | — | 0.0% |
| fill | 205 ms | 219 ms | 100.0% |
| flatten | 158 ms | 167 ms | 100.0% |
| font-inspect | 258 ms | 273 ms | 100.0% |
| get-info | 106 ms | 107 ms | 100.0% |
| grayscale | 158 ms | 168 ms | 100.0% |
| html-to-pdf | 2,388 ms | 2,364 ms | 100.0% |
| image-to-pdf | 107 ms | 131 ms | 100.0% |
| merge | 158 ms | 160 ms | 100.0% |
| nup | 307 ms | 298 ms | 100.0% |
| ocr | 5,608 ms | 8,503 ms | 100.0% |
| office-scan | 33,116 ms | 33,475 ms | 100.0% |
| outline-editor | 256 ms | 269 ms | 100.0% |
| pdf-scan | 4,896 ms | 4,981 ms | 100.0% |
| pdf-to-excel | 1,007 ms | 1,011 ms | 100.0% |
| pdf-to-html | 306 ms | 292 ms | 100.0% |
| pdf-to-md | 1,158 ms | 1,157 ms | 100.0% |
| pdf-to-ppt | 611 ms | 619 ms | 100.0% |
| pdfa | 206 ms | 207 ms | 100.0% |
| pdfx | 206 ms | 209 ms | 100.0% |
| ppt-to-pdf | 1,458 ms | 1,483 ms | 100.0% |
| protect | 106 ms | 108 ms | 99.3% |
| redact | 358 ms | 376 ms | 100.0% |
| reorder | 310 ms | 314 ms | 100.0% |
| repair | 105 ms | 116 ms | 100.0% |
| rotate | 308 ms | 313 ms | 100.0% |
| sign | 54 ms | 55 ms | 100.0% |
| split | 418 ms | 430 ms | 100.0% |
| table-json | 1,882 ms | 1,888 ms | 100.0% |
| to-images | 209 ms | 210 ms | 100.0% |
| unlock | 106 ms | 108 ms | 99.0% |
| watermark | 258 ms | 266 ms | 100.0% |
| word-to-pdf | 1,758 ms | 1,746 ms | 100.0% |
| workflow | 155 ms | 170 ms | 100.0% |
Cipher: AES-256-GCM
chmod 070024-hex CSPRNG suffix
random_bytes(12)120s OS timeout
unshare + strace
during stream
not after
No retention window
No cleanup cron
| Engine | Version (production) | Used for | Public cap | On-premise |
|---|---|---|---|---|
| Ghostscript | 10.02.1 | Compress, watermark, rotate, protect, unlock, flatten, grayscale, repair, PDF/A, PDF/X | 50 MBthrottle | No limit |
| Poppler (pdfunite, pdftoppm, pdftotext) | bundled | Merge, split, to-images, extract-text, PDF info, page operations | 50 MBthrottle | No limit |
| LibreOffice | 24.2.7.2 | Word/Excel/PowerPoint ↔ PDF conversion | 50 MBthrottle | No limit |
| Tesseract | 5.3.4 | OCR (30+ languages, LSTM), searchable PDF generation, camera scan | 50 MBthrottle | No limit |
| qpdf | 11.9.0 | Structural repair, stream compression, cross-reference validation (forensics) | 50 MB / 10 MB scan | No limit |
| PyMuPDF (fitz) | 1.27.2 | Forensic parsing, page rendering, redaction, metadata, image extraction | 10 MB scan | No limit |
| ClamAV | 1.4.3 | Malware signature scanning (700K+ signatures, updated continuously) | 10 MB scan | No limit |
| YARA | 4.5.0 | Byte-pattern rule matching, custom threat signatures | 10 MB scan | No limit |
| ExifTool | 12.76 | Metadata extraction, author/creator forensics, GPS/timestamp analysis | 10 MB scan | No limit |
| scikit-learn | 1.8.0 | ML anomaly detection — IsolationForest + RandomForest in forensic scanner | 10 MB scan | No limit |
| LightGBM | 4.6.0 | Gradient-boosted ML scoring + SHAP explainability in forensic scanner | 10 MB scan | No limit |
| pikepdf | 10.5.1 | Low-level PDF object manipulation, sign/redact/edit workflows | 50 MBthrottle | No limit |
| Acorn (Node.js) | 8.15.0 | JavaScript AST parsing and deobfuscation in forensic scanner | 10 MB scan | No limit |
| Qwen 2.5 1.5B (llama.cpp) | self-hosted | AI forensic report generation — WireGuard-isolated, no external API calls | 10 MB scan | Included |
The public file size caps are deliberate throttles on pqpdf.com, not engine constraints. Ghostscript, Poppler, and LibreOffice process files of arbitrary size. The forensic scanner's 10 MB limit is tuned to the public threat profile — 99.7% of real-world malicious PDFs are under 5 MB. On your own deployment, both limits are removed.
Every request gets a unique temp directory: /tmp/[tool]-[24-hex-random]/ with permissions 0700. The random suffix is generated with random_bytes(12) — 96 bits of entropy. No two requests share a directory; cross-request data leakage is structurally impossible.
All file paths and user-controlled values passed to shell commands are wrapped with PHP's escapeshellarg(). No user-controlled string ever reaches the shell interpreter unescaped. Command construction is auditable in source.
All engine calls run through PHP's proc_open() or equivalent with a hard 120-second wall-clock timeout enforced at the OS level. A malicious PDF designed to cause infinite processing loops (zip bombs, recursive structures) cannot stall the server.
unshare + strace)The forensic scanner's behavioural analysis engine runs PDF rendering inside a Linux network namespace (unshare --net) to block outbound C2 callbacks, while strace records all system calls. Files that attempt network connections during rendering are flagged immediately.
Deployment & Licensing
A one-time setup and configuration fee covers installation, environment hardening, and handover. An annual support and maintenance licence covers security patches, dependency updates, new tool releases, and direct support. Unlimited users, servers, and usage. Tell us your situation and we’ll scope it. You’ll have a clear picture of what’s involved and what it costs before any commitment.
References
- IBM Security & Ponemon Institute. Cost of a Data Breach Report 2024. Global average: $4.88M; Healthcare average: $9.77M. Published July 2024. Available at ibm.com/reports/data-breach.
- European Parliament. Regulation (EU) 2016/679 (GDPR), Article 83(5). Maximum fine: €20,000,000 or 4% of total worldwide annual turnover for the preceding financial year, whichever is higher. Official Journal of the European Union, 4 May 2016.
- National Institute of Standards and Technology (NIST). Post-Quantum Cryptography Standards. FIPS 203 (ML-KEM), FIPS 204 (ML-DSA), FIPS 205 (SLH-DSA) finalised 13 August 2024. Available at nist.gov/pqcrypto.
- Post-quantum encryption is not a feature of mainstream cloud PDF tools (Adobe Acrobat, Smallpdf, iLovePDF, DocHub) as of March 2026. This reflects independent review of publicly available feature documentation for each product. Verify current feature sets directly with each vendor before making procurement decisions.
- Emsisoft Malware Lab. Unpacking the MOVEit Breach: Statistics and Analysis. Tracked 2,559+ confirmed victim organisations and 66M+ individuals as of October 2023; later tallies reached ~2,600–2,773 organisations and 93–96 million individuals. Progress Software disclosed in SEC filings (2024) that it “cannot reasonably estimate a range of possible losses” from consolidated class-action litigation. SEC declined enforcement action against Progress Software (2024).
- UnitedHealth Group. Full-Year 2024 Earnings Release (January 16, 2025, filed with SEC via BusinessWire). $2.2 billion direct response costs; $3.1 billion total cyberattack impact. HHS Office for Civil Rights: 190 million individuals affected (updated January 2025). Prior quarterly figures: $872M (Q1 2024 10-Q), ~$2.3–2.45B full-year estimate (Q2 guidance), ~$2.9B (Q3 10-Q). No regulatory fine finalised as of March 2026.
- Federal Trade Commission. Equifax to Pay $575 Million as Part of Settlement with FTC, CFPB, and States Related to 2017 Data Breach. Press release, July 22, 2019. Consent order entered U.S. District Court for the Northern District of Georgia. $425M consumer fund; $175M to 48 states, D.C., and Puerto Rico; $100M CFPB civil penalty. Maximum ceiling $700M. Available at ftc.gov.
- Office of the Comptroller of the Currency. OCC Consent Order #2020-134, August 2020 ($80M civil money penalty). U.S. District Court for the Eastern District of Virginia. Capital One class action final settlement approval, 2022 ($190M). 98 million U.S. consumers; 3 years identity theft protection included.
- Information Commissioner's Office (UK). Final Penalty Notice: British Airways plc, October 16, 2020. £20,000,000 imposed under UK GDPR. Initial notice of intent (July 2019) was £183,390,000; final imposed fine was £20M after mitigating factors including cooperation, remediation, and COVID-19 financial hardship reduction. Available at ico.org.uk.
- OCC Consent Order #2020-134 / OCC Enforcement Action EA2020-058 (October 8, 2020): $60M civil money penalty. SEC Enforcement Action #2022-168 (September 20, 2022): $35M penalty for Regulation S-P Safeguards Rule and Disposal Rule violations; 15 million customers' PII on unaccounted devices (sec.gov/newsroom/press-releases/2022-168). Federal class action settlement: $60M (2022, resolving suits from 2016 and 2019 incidents — BankInfoSecurity; Top Class Actions). New York AG multistate settlement: $6.5M (2023, ag.ny.gov press release). Total across four actions: $155M+.
- Check Point Research. The Weaponization of PDFs: 68% of Cyberattacks Begin in Your Inbox, with 22% Hiding in PDFs. blog.checkpoint.com/research, April 2025. Based on Check Point's own email scanning telemetry. Also cites 87% of organisations use PDFs for business and 400 billion+ PDFs opened globally per year.
- HP Wolf Security. Threat Insights Reports: Q1 2024 (May 2024), September 2024, December 2025. threatresearch.ext.hp.com. PDFs accounted for 11% of threats in Q1 2024 and Q3 2025 endpoint telemetry. Q4 2025 report: at least 14% of email threats bypass gateway scanners. September 2025 press release: ultra-realistic AI-generated PDF invoice lures documented in active campaigns.
- Kaspersky. Security Bulletin 2024: Statistics. securelist.com/ksb-2024-statistics, December 2024. Press release: “The Cyber Surge: Kaspersky detected 467,000 malicious files daily in 2024.” kaspersky.com/about/press-releases.
- CVE-2023-26369: Adobe Security Bulletin APSB23-34; Google Project Zero 0days-in-the-wild (North Korean APT confirmation); CISA KEV added September 2023 (cisa.gov/known-exploited-vulnerabilities-catalog). CVE-2023-21608: NVD nvd.nist.gov/vuln/detail/cve-2023-21608; CISA KEV added October 2023; The Hacker News (October 2023). Both vulnerabilities mandated remediation for US federal agencies under BOD 22-01.
- Check Point Research. Foxit PDF Flawed Design Exploitation. research.checkpoint.com, May 2024. Confirmed multiple threat actor groups exploiting Foxit PDF Reader logic flaw to deliver Agent Tesla, Remcos RAT, NanoCore, and ransomware droppers. CIS Multi-State Advisory issued (cisecurity.org, 2024-105). The Hacker News: “Foxit PDF Reader Flaw Exploited by Multiple Threat Actors,” May 2024.
- EXPMON (justhaifei1.blogspot.com). Zero-Day Behavior in Adobe Acrobat and Foxit PDF Reader — NTLM Leak. December 2024. Both Adobe and Foxit confirmed the behaviour and issued patches after responsible disclosure. cybersecuritynews.com: “Zero-Day Vulnerability in PDF Files Leaking NTLM Data in Adobe & Foxit Reader,” December 2024. No user interaction required beyond opening the PDF.
- Resecurity. PDFSider Malware: Exploitation of DLL Side-Loading for AV and EDR Evasion. resecurity.com/blog, January 2026. BleepingComputer: “New PDFSider Windows Malware Deployed on Fortune 100 Firm's Network,” January 2026. SecurityWeek: “APT-Grade PDFSider Malware Used by Ransomware Groups,” January 2026. HP Wolf Security press release (September 2025): AI-generated PDF invoice lures documented in active campaigns.
- Anthem $115M class action: U.S. District Judge Lucy Koh, N.D. Cal., Case No. 15-md-02617, final approval August 15, 2018 (Hunton Andrews Kurth; HIPAA Journal). Anthem AG settlements ($48.2M): NY AG press release (ag.ny.gov, October 2020) — $39.5M multistate (43 states + DC, led by Connecticut AG); California AG press release (oag.ca.gov) — $8.69M. Anthem HHS OCR HIPAA settlement: $16M, October 15, 2018, record HIPAA penalty at the time (hhs.gov/hipaa/for-professionals/compliance-enforcement/agreements/anthem). Total across three enforcement actions: $179.2M+. 78.8 million records. 2014 breach.
- Verizon. 2025 Data Breach Investigations Report. Third-party involvement in breaches: 30% (doubled year-over-year from 15% in the 2024 DBIR). Ransomware in 44% of confirmed breaches (up from 32%). Phishing in 16% of breaches; human element in 60%. verizon.com/business/resources/reports/dbir/; Executive Summary PDF: verizon.com/business/resources/reports/2025-dbir-executive-summary.pdf. Verizon 2024 DBIR (prior year): supply chain attacks 15% of all breaches, 68% YoY increase. Verizon DBIR 2024: phishing in 14% of credential-related breaches; median click time under 60 seconds.
- Irish Data Protection Commission. Final Decision: Meta Platforms Ireland Limited. Adopted May 12, 2023. Fine: €1,200,000,000 — the largest GDPR fine ever imposed as of March 2026. Violation: Article 46(1) GDPR, unlawful transfers of EU/EEA personal data to the US post-Schrems II. Implementing EDPB Binding Decision 1/2023 (adopted April 13, 2023, Article 65 dispute resolution). Official DPC source: dataprotection.ie/en/news-media/press-releases. EDPB source: edpb.europa.eu/news/news/2023/12-billion-euro-fine-facebook-result-edpb-binding-decision_en.
- GoAnywhere MFT / Fortra: CVE-2023-0669 exploited by Cl0p beginning January 25, 2023; ~130 victim organisations. Fortra $20M MDL settlement: final approval confirmed (HIPAA Journal; classaction.org). Brightline $7M settlement: final approval February 2025, S.D. Fla. (TechTarget HealthTech Security; brightlinedatasecuritysettlement.com). Named victims include Hitachi Energy, P&G, Rubrik, City of Toronto, Saks Fifth Avenue, Rio Tinto, UK Pension Protection Fund.
- Accellion FTA: multiple zero-days exploited by Cl0p; ~100 of ~300 FTA customers breached; 9.2M+ individuals. Accellion $8.1M class action settlement ($4.6M + $3.5M tranches) — SecurityWeek; HIPAA Journal; TechTarget. Kroger $5M settlement (July 2021) for 1,474,284 patient pharmacy records — ComplianceJunction. Confirmed victims: ASIC (Australia), Bombardier, Royal Dutch Shell, Kroger, Flagstar Bank, universities. Flagstar Bank separately received a $3.5M SEC fine (December 2024) for making “materially misleading statements” about this breach in its 2021 Form 10-K (Cybersecurity Dive; SEC enforcement order).