Cybersecurity for SaaS: What the Biggest SaaS Breaches Reveal About Securing Cloud Applications
Executive Summary: SaaS security failures rarely come from broken vendor infrastructure. The largest SaaS breaches of the past three years, including Snowflake, Salesloft Drift, Microsoft’s Midnight Blizzard intrusion, Okta’s support system compromise, and CircleCI, all came through stolen credentials, hijacked sessions, or over-privileged OAuth tokens. Under the shared responsibility model, those identity and configuration layers belong to the customer. Organizations that enforce phishing-resistant MFA, govern non-human identities, and monitor SaaS configurations close the gaps attackers have used most often.
SaaS has become the default way businesses run email, CRM, file storage, development pipelines, and data warehouses. The security model has not kept pace with that adoption.
Most organizations still treat SaaS security as the vendor’s problem. The breach record says otherwise. In case after case, attackers walked in through doors the customer controlled: unprotected accounts, stale credentials, and third-party integrations nobody was watching.
This article covers what cybersecurity for SaaS actually involves, what five major incidents show about where SaaS defenses fail, and which controls address those failure points.
What “Cybersecurity SaaS” Means
The term carries two meanings, and buyers searching for it usually want one of them.
The first is SaaS security: protecting the software-as-a-service applications an organization uses, along with the data stored inside them. That covers access control, configuration management, integration governance, and threat detection across platforms such as Microsoft 365, Salesforce, Google Workspace, and Snowflake.
The second is security delivered as SaaS. Endpoint detection, SIEM, identity providers, and email security are now commonly sold as cloud-hosted subscriptions rather than on-premises software.
The two overlap. A cloud-delivered identity provider protects other SaaS applications, and it is also a SaaS application that needs protecting. Okta’s 2023 incident, covered below, shows what happens when that layer is compromised.
This article focuses on the first meaning: securing the SaaS environments an organization depends on.
The Shared Responsibility Gap
SaaS providers secure the infrastructure, platform, and application code. Customers secure everything they configure and control.
| Layer | SaaS vendor responsibility | Customer responsibility |
|---|---|---|
| Physical infrastructure | Data centers, hardware, network | None |
| Application code | Patching, vulnerability management | None |
| Identity and access | Providing authentication options | Enforcing MFA, managing users, revoking access |
| Configuration | Providing security settings | Configuring them correctly |
| Integrations | Offering OAuth and API frameworks | Approving, scoping, and auditing connected apps |
| Data | Platform-level encryption | Classification, sharing controls, retention |
The customer side of that table is where most SaaS breaches begin. A vendor can offer MFA, network allow lists, and granular permissions. None of those controls protect anything until a customer turns them on.
The SaaS attack surface also grows quietly. Each new application, integration, service account, and guest user adds exposure. Security teams that practice attack surface management for on-premises assets often lack the same inventory for SaaS.
Five SaaS Breaches and What They Exposed
The following incidents span data platforms, sales tools, identity providers, and developer infrastructure. Each followed a different path in. The weaknesses underneath were strikingly similar.
Snowflake (2024): Stolen Credentials, No MFA
In mid-2024, Mandiant reported that a financially motivated threat actor it tracks as UNC5537 had systematically accessed Snowflake customer instances using stolen credentials. Mandiant and Snowflake notified approximately 165 potentially exposed organizations.
The credentials came from infostealer malware on systems Snowflake did not own. Mandiant said many of those infections dated back as far as 2020.
Snowflake’s own platform was not breached. Mandiant’s analysis pointed to the customer side: the affected instances did not require MFA, many credentials had not been rotated in as long as four years, and network allow lists were not in place to restrict access to trusted locations. Mandiant also stressed that the campaign used no particularly novel or sophisticated technique.
What it exposed: Credentials harvested years earlier remained valid because nothing forced rotation, and no second factor stood in the way. This is credential stuffing at enterprise scale.
Controls that would have helped: Enforced MFA at the tenant level, network policies limiting access to known IP ranges, and credential rotation tied to exposure monitoring.
Salesloft Drift (2025): OAuth Tokens as the Attack Path
In August 2025, Google Threat Intelligence Group (GTIG) warned of a widespread data theft campaign by a threat actor it tracks as UNC6395. Between August 8 and August 18, the actor used compromised OAuth tokens tied to Salesloft’s Drift AI chat integration to access Salesforce customer instances. According to GTIG’s advisory, the actor systematically exported large volumes of data and searched it for secrets such as AWS access keys and Snowflake tokens.
Google told Cybersecurity Dive it was aware of more than 700 potentially impacted organizations. On August 9, the actor also used tokens from the Drift Email integration to access a small number of Google Workspace accounts.
Salesloft and Salesforce revoked all active Drift access and refresh tokens on August 20. Google later advised Drift customers to treat every authentication token stored in or connected to the platform as potentially compromised.
Salesforce itself was not breached. The attackers never needed a password or an MFA prompt. They held tokens that the integration had already been granted.
What it exposed: Third-party integrations carry standing access that bypasses user authentication entirely. Data also became a launchpad, because secrets stored in CRM records gave attackers keys to other environments.
Controls that would have helped: Least-privilege OAuth scopes, an inventory of connected apps, monitoring of API query volume, and keeping secrets out of SaaS records.
Microsoft Midnight Blizzard (2024): A Forgotten Test Tenant
In January 2024, Microsoft disclosed that the Russian state-sponsored group Midnight Blizzard, also tracked as APT29, had accessed email accounts belonging to members of its senior leadership team.
According to Microsoft, initial access came through a password spray attack against a legacy, non-production test tenant account that did not have MFA enabled. The attackers used residential proxies and a low volume of attempts to avoid triggering brute-force detections.
From there, the group compromised a legacy test OAuth application that had elevated access to Microsoft’s corporate environment. The attackers used it to grant themselves the Office 365 Exchange Online full_access_as_app role, which provides access to mailboxes.
What it exposed: Non-production environments and legacy applications often escape the security standards applied to production. An over-privileged test app became a bridge into the corporate tenant, a textbook case of lateral movement through identity rather than network access.
Controls that would have helped: MFA on every account regardless of environment, regular audits of OAuth application permissions, and decommissioning of legacy test assets.
Okta (2023): Session Tokens in Support Files
In October 2023, Okta disclosed that a threat actor had used a stolen credential to access its customer support case management system. Okta later reported that between September 28 and October 17, the attacker accessed files associated with 134 customers.
Some of those files were HTTP Archive (HAR) files that customers had uploaded to troubleshoot problems. HAR files record browser sessions, and those included valid session tokens. Okta said the attacker used them to hijack the legitimate sessions of five customers.
A follow-up review found the attacker had also downloaded a report containing the names and email addresses of all users of the affected support system.
What it exposed: Session tokens sidestep MFA. A captured token represents an already-authenticated session. Routine operational processes such as support ticketing can also leak authentication material.
Controls that would have helped: Sanitizing HAR files before upload, short session lifetimes, session binding to devices, and alerting on session reuse from new locations.
CircleCI (2023): Malware Defeats 2FA
In January 2023, CI/CD provider CircleCI published an incident report explaining a breach that forced customers to rotate every secret stored on its platform.
According to the report, an unauthorized third party deployed malware on an engineer’s laptop and stole a valid, 2FA-backed SSO session. The company said its antivirus software did not detect the malware.
The engineer had privileges to generate production access tokens. Using the stolen session, the attacker accessed a subset of production systems and exfiltrated data that included customer environment variables, tokens, and keys.
What it exposed: MFA protects the login. It does not protect the session after login. A single compromised endpoint with privileged access put the secrets of an entire customer base at risk.
Controls that would have helped: Stronger endpoint detection and response, shorter-lived sessions, re-authentication for privileged actions, and stricter limits on who can generate production tokens.
The Pattern Across All Five
| Incident | Entry point | Underlying weakness |
|---|---|---|
| Snowflake | Infostealer-harvested passwords | No MFA, no credential rotation |
| Salesloft Drift | Stolen OAuth tokens | Over-trusted third-party integration |
| Midnight Blizzard | Password spray on test account | No MFA, over-privileged legacy app |
| Okta | Session tokens in support files | Session tokens bypass MFA |
| CircleCI | Malware-stolen SSO session | Endpoint compromise, privileged sessions |
None of these incidents required a zero-day exploit in the SaaS platform itself. All five came down to identity: who or what could authenticate, and what it could reach afterward.
Identity Is the SaaS Perimeter
SaaS applications sit on the public internet. There is no network edge in front of them. Identity is the control point.
Enforce MFA Everywhere, Then Go Further
Snowflake and Midnight Blizzard show the cost of accounts without MFA. Every human account, in every environment, needs a second factor.
MFA strength varies, though. SMS codes and push approvals can be phished or worn down by prompt fatigue. Phishing-resistant methods, such as FIDO2 security keys and passkeys, bind authentication to the legitimate site and cannot be relayed through a fake login page.
Prioritize phishing-resistant MFA for administrators, developers, and anyone with access to production or sensitive data.
Treat Sessions as Credentials
The Okta and CircleCI incidents both bypassed MFA by stealing sessions. Authentication controls end at login. Session controls have to cover everything afterward.
Practical session controls include:
- Shorter session lifetimes for privileged access
- Re-authentication before sensitive actions
- Device-bound sessions where the platform supports them
- Alerts on session reuse from new IP addresses or locations
Centralize Through SSO
Single sign-on gives security teams one place to enforce MFA, apply conditional access, and revoke access quickly. Applications outside SSO are blind spots. Inventory them and bring them under central control where possible.
This approach aligns with zero trust principles: verify every access request, regardless of where it originates.
Non-Human Identities: The Overlooked Access Layer
Service accounts, API keys, and OAuth tokens now outnumber human users in many SaaS environments. They often carry broad permissions, rarely expire, and sit outside MFA entirely.
Salesloft Drift and Midnight Blizzard both turned on non-human identities. In each case, an application held more access than its function required.
Governing non-human identities starts with visibility:
- Inventory every connected app. Most SaaS admin consoles list authorized OAuth applications. Review them regularly.
- Scope permissions to function. A chat integration should not have read access to an entire CRM.
- Rotate and expire. Set expiration on API keys and tokens. Revoke credentials tied to unused integrations.
- Monitor behavior. Unusual query volume, new IP ranges, or bulk exports from an integration account warrant investigation.
- Keep secrets out of SaaS data. Credentials stored in CRM notes, tickets, or chat messages become targets once an attacker gets read access.
Misconfiguration, Sharing, and Offboarding
Identity attacks get the headlines. Day-to-day SaaS exposure more often comes from settings that drift over time.
Misconfigurations
SaaS platforms ship with hundreds of security settings. Defaults often favor usability over security, and settings change as administrators troubleshoot, onboard users, and add features. Without continuous monitoring, an organization’s security posture degrades without anyone noticing.
External Sharing and Public Links
File-sharing features create persistent exposure. Links set to “anyone with the link” survive project completion, staff changes, and client offboarding. Agencies and creative teams that send large files to clients face this risk constantly.
Effective controls include:
- Default sharing restricted to named recipients
- Expiration dates on external links
- Periodic audits of publicly shared files
- Watermarking or view-only access for sensitive work in progress
Offboarding
Departing employees, contractors, and freelancers often keep SaaS access long after they leave. The risk compounds when they connected personal accounts or authorized third-party apps under their own credentials.
Offboarding needs to revoke more than the login. It should also cover OAuth grants the user authorized, API keys they created, and files they shared externally. Centralizing access through SSO makes this far more reliable.
SaaS Security Tool Categories
Several product categories address the SaaS attack surface. Most organizations combine more than one.
| Category | What it does | Addresses |
|---|---|---|
| Identity provider (IdP) / SSO | Centralizes authentication and access policy | Account takeover, inconsistent MFA |
| SaaS security posture management (SSPM) | Monitors SaaS configurations against baselines | Misconfiguration and drift |
| Cloud access security broker (CASB) | Monitors and controls SaaS usage and data movement | Shadow IT, risky sharing |
| Identity threat detection and response (ITDR) | Detects identity-based attacks in progress | Session hijacking, token abuse |
| Data loss prevention (DLP) | Identifies and blocks sensitive data movement | Data exfiltration, oversharing |
| SIEM / XDR | Correlates logs and alerts across systems | Detection and incident response |
Tooling depends on logging. Confirm that each SaaS platform’s audit logs are enabled, retained long enough to support investigations, and forwarded to a central system. Several of the incidents above were reconstructed only because logs existed.
Evaluating a SaaS Vendor’s Security
Vendor due diligence reduces risk before data ever reaches a platform. Key questions include:
- Authentication: Does the platform support SSO, enforced MFA, and phishing-resistant methods?
- Access controls: Can permissions be scoped by role? Can admins restrict access by IP address or device?
- Integrations: Can admins see and revoke every connected OAuth application?
- Logging: What audit logs are available, how long are they retained, and can they be exported?
- Attestations: Does the vendor provide a current SOC 2 Type II report or equivalent independent assessment?
- Data handling: Where is data stored? Is customer data used to train AI models?
- Incident history: How did the vendor handle past incidents, and how quickly did it notify customers?
A vendor’s security features matter only if they are enabled. Build configuration requirements into onboarding, not into the post-incident review.
Frequently Asked Questions
What is SaaS in cybersecurity?
In cybersecurity, SaaS refers either to securing cloud-hosted applications an organization uses or to security tools delivered as cloud subscriptions. SaaS security covers access control, configuration, integrations, and data protection within those applications.
Who is responsible for SaaS security?
Responsibility is shared. The vendor secures the infrastructure and application. The customer secures identities, configurations, integrations, and data.
What is the most common cause of SaaS breaches?
Identity compromise. Stolen credentials, hijacked sessions, and over-privileged tokens account for most major SaaS incidents.
Do organizations using SaaS security tools still need penetration testing?
Yes. Tools monitor known configurations and behaviors. Penetration testing and attack simulation find gaps tools may not cover, including integration abuse paths and privilege escalation.
The Breach Record Already Wrote the Playbook
The five incidents above share almost no infrastructure. They share nearly all of their lessons.
Attackers did not need to break SaaS platforms. They used what customers left open: accounts without MFA, sessions without limits, tokens without scope, and test environments without oversight.
That is encouraging, in a way. Every one of those gaps falls on the customer’s side of the shared responsibility model, which means every one of them is fixable.
Key Takeaways
- Most major SaaS breaches stem from identity compromise, not vendor infrastructure failures.
- Enforce MFA on every account in every environment, with phishing-resistant methods for privileged users.
- Treat session tokens as credentials: limit lifetimes, require re-authentication for sensitive actions, and monitor reuse.
- Inventory and scope OAuth applications, API keys, and service accounts; revoke anything unused.
- Monitor SaaS configurations continuously to catch misconfiguration and drift.
- Audit external sharing and fully revoke access, including OAuth grants, during offboarding.
- Verify vendor security capabilities before onboarding, then confirm they are actually enabled.