Informational

Wrangling your SaaS Providers: Managing Your Rented Risk

How SaaS compromise actually happens, the obligations orgs retain under SOCI and the Privacy Act, and how to move from point-in-time vendor assessment to continuous verification

Authored By

David Sandell

Data published

November 16, 2026

EXECUTIVE SUMMARY

In a SaaS environment you rent the application, keep the function, and pay a bill for the service. Your security team sleeps soundly having locked down the tenant, but behind the application sits the SaaS provider’s own supply chain: hyperscalers, integration partners, offshore engineers, support desks, and AI models. You inherit all of that exposure with less visibility and less contractual leverage than in almost any other business arrangement.

The larger problem is that the vendor you assessed at procurement isn’t the vendor you’re with today. Since you signed, it has changed some combination of who writes code, who’s on the helpdesk, which subprocessors touch your data, and which AI features exist in the environment; and your contract allowed all of it. Insurers have abandoned self-assessment questionnaires because they become stale too quickly. Vendor security questionnaires carry the same flaw, yet most of us still rely on them.

So what do we do about it? We outline how SaaS compromise actually happens, the obligations members retain under SOCI and the Privacy Act, and how to move from point-in-time vendor assessment to continuous verification. When any of it fails, the regulator holds you accountable, not the vendor.

THREAT CONTEXT

The enterprise attack surface has shifted from the network perimeter to SaaS platforms carrying IDAM, collaboration, and core business processes. How do you manage something you don't own, can't inspect, and only see through jargon-filled contracts, a well-funded marketing team, and a battalion of relationship or customer success managers between you and the workers? The SaaS vendor runs the application, though you’re still responsible for running and securing the function, on code and infrastructure that you don’t own, and that is maintained and protected by phantom teams you’ve never met or vetted, and those teams are shrinking.

Core problems

Heads are rolling, fewer people and more vulns

SaaS vendors are rolling heads like every other business, swapping people for cheaper overheads, reducing engineering, assurance, tech, and support roles while plugging the gaps with AI. For example, Monday.com and Wix cut about 20% of their staff as they pivot to AI agents, Oracle cut roughly 21,000 people early 2026 ‘embracing AI’, Microsoft is obliterating 60,000 workers, more than 20% of their global workforce. Amazon cut 16,000 in January 2026 on top of 14,000 late 2025, while Workday just announced another 500 cuts in their second round of layoffs in 2026 - the list goes on and on. Why care who your vendor lays off? Because those people are part of your security team, they review the code, patch the bugs, and answer the support calls.

All of this landing while vulnerability disclosure soars

As of late October 86,926 CVEs have been published in 2026 - roughly 290 a day, up from 131 a day in 2025, and AI-assisted discovery is a big part of the jump. For a vendor that’s just cut half its team this means a growing backlog to triage, fix, and test which is now landing on fewer people. Something has to give... it's usually the patch speed and quality. Security teams aren’t the only ones paying attention, threat actors have noticed too, and they’re cashing in. Verizon’s 2026 DBIR found vulnerability exploitation now accounts for 31% of initial access, well ahead of credential abuse at 13%, and IBM saw a 44% rise in exploitation of public-facing applications. Threat actors take the path of least resistance, and SaaS raises those stakes significantly as a vulnerability in a multi-tenant platform exposes every customer running it.

Your vendor’s supply chain is your supply chain

You might call your org an Azure shop but if your SaaS vendor runs on AWS, that label stops meaning much, you’re an AWS shop too. Your attack surface is everything your vendors depend on, from CDNs and identity providers to support platforms and bolted-on integrations, and your threat model should be covering all of it. We’ve covered the software supply chain at length, and 2026 has seen a long tail of SaaS vendors in the firing line. Who has access to my data?

Data sovereignty and provenance

Vendors slapping on “Hosted in Australia” has lost its meaning and only really answers one of five important questions our members need to consider - what about backups, support access, telemetry, and where integrated AI features run? Any of those can and do sit offshore, even when the ‘main’ data is in Australia. The people are heading offshore too, Revelio Labs found offshore headcount at US tech companies grew from 2019 to 2024 by 78% for software engineers, 62% for infosec, and 61% for DevOps. The problem isn’t necessarily the firings, it’s that this changes your risk profile, and chances are the first place you hear about it is on the news. Subprocessor lists are a strong starting point (often available upon request under ISO 27001 and similar frameworks), but they’re a snapshot and vendors don’t always notify customers when they make changes.

Who writes, manages, and ships the code?

AI is now writing the code and a “human signs off”. SaaS vendors are cutting staff and leaning harder on AI to fill the gap resulting in a growing share of the platform members rely on being machine-written. Google confirmed 75% of its new code is AI-generated and approved by engineers, and Microsoft confirmed 20-30% of code in its repositories is written by software. Meanwhile numerous benchmarking results suggest around half of AI-generated code still fails security tests, doesn’t fix bugs, or breaks functionality when trying to fix things. When the volume of machine output outpaces the people reviewing it the human sign-off becomes the only control, and realistically it’s a fake one. No human is reading these 10,000-line pull requests, they’re just running the test suite the AI wrote for itself. Members can’t see how many reviewers a vendor has, how much code each approves, or, whether review happens offshore, ask.

What the product does with your data

Vendors are switching on AI features inside the product, often by default, and adjusting terms to permit training on customer data (Slack and Zoom both faced public backlash over exactly this). Agentic features create exfiltration paths that didn't exist when you procured the platform: EchoLeak (CVE-2025-32711) in Microsoft 365 Copilot and ForcedLeak in Salesforce Agentforce both used prompt injection to pull customer data out through features customers never asked for.

Regulatory Position

Outsourcing the function does not outsource the obligation. SaaS adds some specific points to this:

  • SOCI Risk Management: The CIRMP Rules name supply chain as one of four hazard domains, explicitly covering unauthorised access and misuse of privileged access by vendors, disruption, and over-reliance on particular suppliers. A SaaS platform you cannot inspect, staffed by people you have never vetted, sits squarely inside it.
  • SOCI Incident Reporting: Critical cyber security incidents must be reported within 12 hours of becoming aware, and other relevant incidents within 72 hours. If your only route to awareness of a SaaS compromise is the vendor's notification, the vendor controls when your clock starts and how much of it you have left to investigate. A 72-hour vendor notification clause does not support a 12-hour obligation.
  • APRA CPS 230: For APRA-regulated members, CPS 230 has applied since July 2025, covering material service provider registers, notification to APRA, and credible exit planning. Members outside APRA's remit should borrow its structure; it is the most mature Australian template for this problem.
  • Privacy Act: APP 11 and APP 8 apply to SaaS exactly as our MSP advisory explains, including where offshore support staff or AI features access personal information. This is becoming more pertinent, and we expect one day OAIC will wake up and start punishing MSPs and SaaS vendors for flagrant mishandling of their customers' data, on top of actually policing the current rules that threaten significant financial punishment for the data custodians (that’s you) themselves.

Recommendations

  1. Build a fourth-party register: For each critical SaaS platform, record hosting provider and region, CDN, identity provider and subprocessors. Flag where one provider sits behind several vendors. CI-ISAC offers assistance with mapping your vendor/software dependencies.
  2. Read the assurance report you already have: A SOC 2 report’s Complementary User Entity Controls section lists the controls the vendor assumes you operate. Check what is carved out and make sure you’re doing it yourself.
  3. Build staffing, offshoring, and contractor changes into the contract: Require advance notice of changes to support location, engineering location, contractor or subcontractor use, AI in  development and support, and change of control, with a right to object or exit. Set incident notification timeframes that leave room inside your own SOCI reporting window, and make sure those incidents are defined broadly so they cover things like single-address BEC that often goes unreported.
  4. Plan your exit before you need it: Test a full data export in a usable format and document a realistic migration timeframe for each critical platform. When the contract ends, export and verify what the vendor holds, then get written confirmation of deletion - a signed statement covering production data, backups, snapshots, replicas, logs and support tickets, including all subprocessor copies and any data used to train or tune AI models.
  5. Lock down your tenant: Name an internal owner per platform. Enforce phishing-resistant MFA and allow lists, keep log retention under your control. Inventory every OAuth grant, service principal and API key with its permissions, remove unused ones, set expiry, restrict scopes and alert on new grants and self-assigned roles. Write revocation runbooks per integration. We recommend removing unused apps and features, scope the rest to least privilege, and be ready to revoke and rotate accounts.
  6. Ask for provenance evidence: Request an SBOM from the vendor. ISM-1730 says an SBOM is produced and made available to consumers of software, and the July 2026 ASD-co-authored minimum elements apply to all software including AI and SaaS. Members are encouraged to ask where builds run, who can ship a release, how contractor identity is verified and how AI-generated code is reviewed.

REFERENCES

  • https://www.jpost.com/business-and-innovation/article-903638
  • https://www.bbc.com/news/articles/c4gy0x0j5deo
  • https://ia.acs.org.au/article/2026/microsoft-offers-60-000-workers-voluntary-redu
    ndancies.html
  • https://edition.cnn.com/2026/01/28/tech/amazon-layoffs-ai
  • https://www.latestly.com/business/workday-layoffs-saas-provider-eliminating-500
    -jobs-in-2nd-workforce-reduction-in-2026-7626482.html
  • https://craci.com/resources/CVE-publication-trends
  • https://www.verizon.com/business/resources/reports/2026-dbir-data-breach-inves
    tigations-report.pdf
  • https://www.ibm.com/reports/threat-intelligence
  • https://www.reveliolabs.com/news/business/american-companies-are-offshoring-hi
    gh-paying-remote-friendly-jobs
  • https://offshore-watch.com/company/national-australia-bank
  • https://axis-intelligence.com/ai-generated-code-statistics/
  • https://www.entrepreneur.com/business-news/ai-is-taking-over-coding-at-microso
    ft-google-and-meta/490896
  • https://1password.com/blog/why-ai-generated-patches-still-require-human-review

‍

Related Content

Informational

November 4, 2026

Consumer wearables, the enterprise security threat

AI-enabled consumer wearables have moved beyond far-fetched gadgets into something people use in everyday life, from your local barista to the stranger next to you on the train.

Learn More