Attackers now use AI to write fake emails, copy voices, and fake faces on video calls. Old security tools were not built for this.
- Use AI defenders to handle scale and protect AI agents, identity, workspace, code, and fraud surfaces.
- Pick vendors by risk area, run proofs of concept on real data, and measure false positives and integrations.
- Keep security basics: multi factor login, patching, backups, staff training, and only automate responses after accuracy improves.
At the same time, firms give AI agents access to email, code, and customer data. Each agent is a new door for attackers.
AI cybersecurity companies in San Francisco are building defenses for both problems. This guide covers 10 of them, what they protect, and where each fits.
What Do AI Cybersecurity Companies Do?
These firms use AI to spot and stop threats faster than people can. Many also guard AI tools themselves.
Their work spans several areas:
- Email and workspace security: Stopping fake emails and stolen accounts
- Identity security: Deciding which people and AI agents can reach what
- Security operations: Sorting alerts and speeding up reviews
- Code security: Finding flaws and bad code packages before release
- Fraud and impersonation: Detecting deepfakes, fake accounts, and scams
Why AI Changes the Security Game
AI helps attackers in three big ways.
First, it makes fake emails look perfect. Spelling mistakes used to give scams away. Second, deepfakes let criminals pose as bosses on calls. Third, AI writes harmful code faster.
Defenders need AI too. A security team cannot read millions of emails or alerts by hand. AI handles the volume, and people make the final calls.
How We Picked These Companies
Each company on this list is headquartered in San Francisco. Each one uses AI as a core part of its security product.
We considered:
- What each product protects and how AI powers it
- Funding, scale, and named customers
- Recent launches tied to AI threats or AI agents
- Gaps and risks buyers should consider
Most figures come from company announcements and funding reports. Validate claims in a proof of concept.
Quick Comparison: AI Cybersecurity Companies in SF
| Company | Security Area | AI Strength | Best For |
|---|---|---|---|
| Abnormal AI | Email and collaboration | Behavior-based threat detection | Enterprises facing phishing |
| Material Security | Google Workspace and Microsoft 365 | Protects email, files, and accounts | Cloud-first companies |
| Okta | Identity | Identity security for AI agents | Organizations deploying agents |
| Persona | Identity verification | Deepfake and spoof detection | Platforms verifying real users |
| Exaforce | Security operations | Agentic SOC platform | Teams overwhelmed by alerts |
| Semgrep | Code security | AI triage of code findings | Engineering and AppSec teams |
| Socket | Software supply chain | Detects malicious open source | Teams using open source heavily |
| Doppel | Social engineering defense | Finds and removes impersonations | Brands targeted by scams |
| Sardine | Fraud prevention | Fraud and compliance agents | Fintechs and payment platforms |
| Coalition | Cyber insurance plus security | Risk monitoring tied to coverage | Small and mid-sized businesses |
The 10 Best AI Cybersecurity Companies in San Francisco
1. Abnormal AI: Stopping Email Attacks With Behavior AI
Abnormal AI is based in San Francisco. Its AI learns how people normally write and talk, then flags messages that do not fit.
That catches attacks with no bad links or files, like a fake bill from a “known” vendor. Abnormal also guards Slack, Teams, Zoom, and apps like Salesforce and Workday.
Abnormal was valued at $5.1 billion in 2024. It now serves more than 4,500 organizations, including over a quarter of the Fortune 500.
Best for: Enterprises facing business email compromise and vendor fraud.
Watch out for: Email security works best alongside strong identity controls, not instead of them.
2. Material Security: Protecting Workspace Data
Material Security was founded in 2017 and is headquartered in San Francisco. It protects email, files, and accounts inside Google Workspace and Microsoft 365.
Material spots fake emails, finds private files in shared drives, and locks down stolen accounts. One of its agents cuts off risky links to outside apps and AI agents. Customers include OpenAI, Figma, Mars, and DoorDash.
Material raised a $100 million Series C in 2022, led by Founders Fund, at a $1.1 billion valuation.
Best for: Cloud-first companies running on Google Workspace or Microsoft 365.
Watch out for: It focuses on workspace tools. You will still need endpoint and network security.
3. Okta: Identity Security for AI Agents
Okta is a public company based in San Francisco. It controls how people and systems sign in to work apps.
In March 2026, Okta shared a plan for securing AI agents. Its Okta for AI Agents product gives each agent an identity and a human owner. Teams can list agents, set their access, and cut it off at once if something looks wrong.
Okta’s network includes more than 8,200 app integrations.
Best for: Organizations rolling out AI agents across many business apps.
Watch out for: Identity tools need careful setup. Plan role design and access reviews.
4. Persona: Verifying Real People in the Age of Deepfakes
Persona is a San Francisco firm that checks identities online. It helps sites confirm users are real, old enough, and who they say they are.
Persona has detected more than 75 million AI-driven face spoof attempts. Customers include OpenAI, LinkedIn, Etsy, and Block. It also partners with Okta to verify workforce identities.
In April 2025, Persona raised $200 million at a $2 billion valuation, co-led by Founders Fund and Ribbit Capital.
Best for: Marketplaces, fintechs, and AI platforms that must verify real users.
Watch out for: Verification adds friction. Balance security with smooth user sign-up.
5. Exaforce: An Agentic Security Operations Center
Exaforce was founded in 2022 and is based in San Francisco. It builds an AI-run security center that watches for attacks.
Its AI agents, called Exabots, spot, sort, and look into threats, then help respond. Its data platform aims to replace older log tools called SIEMs.
Exaforce raised a $125 million Series B in May 2026.
Best for: Security teams buried in alerts and short on analysts.
Watch out for: Replacing a SIEM is a big project. Run both in parallel during the switch.
6. Semgrep: AI That Triages Code Security Findings
Semgrep was founded in 2017 and is headquartered in San Francisco. Its open-source engine scans code for security flaws.
Semgrep Assistant uses AI to filter out false alarms and suggest fixes. It can even propose a code change for review. It also lets AI coding tools scan the code they write.
Semgrep raised a $100 million Series D in February 2025, led by Menlo Ventures.
Best for: Engineering teams shipping lots of code, including AI-written code.
Watch out for: AI triage reduces noise but does not replace security review of critical code.
7. Socket: Catching Malicious Open-Source Packages
Socket is a software supply chain security company. It checks open-source packages before they enter your codebase.
Socket uses AI plus human review to catch harmful code and real risks. Customers include Anthropic, Cursor, Figma, Vercel, Replit, and Gusto.
In May 2026, Socket raised a $60 million Series C at a $1 billion valuation, led by Thrive Capital.
Best for: Software teams that rely heavily on open-source libraries.
Watch out for: Blocking packages can slow developers. Tune policies to balance speed and safety.
8. Doppel: Fighting Impersonation and Deepfakes
Doppel fights scams that trick people. It finds fake sites, accounts, and ads that copy a brand, then gets them taken down.
Doppel also trains employees using deepfake simulations, so they learn to spot fake calls and videos. Customers include Coinbase, Ramp, and law firm Orrick.
In November 2025, Doppel raised a $70 million Series C led by Bessemer Venture Partners, at a valuation above $600 million.
Best for: Brands and financial firms targeted by scams and impersonation.
Watch out for: Takedowns depend on platforms responding. Some removals take time.
9. Sardine: AI for Fraud and Compliance
Sardine is a fraud prevention company serving more than 300 enterprises in 70 countries. Customers include FIS, Deel, and GoDaddy.
Sardine raised a $70 million Series C led by Activant Capital. It is building AI agents for fraud and rule checks. These agents help check new customers, screen names, rate merchants, and handle payment disputes.
That combination helps fintechs reduce fraud losses without slowing good customers.
Best for: Fintechs, payment companies, and marketplaces fighting fraud.
Watch out for: Fraud models need tuning to your business. Expect a calibration period.
10. Coalition: Cyber Insurance With Built-In Security
Coalition was founded in 2017 and is headquartered in San Francisco. It combines cyber insurance with security technology.
Coalition watches each customer’s outside risks and warns them to fix problems before attackers strike. It also covers some AI risks. This suits firms without big security teams.
Coalition serves businesses, insurance brokers, and security teams.
Best for: Small and mid-sized businesses wanting insurance and risk guidance together.
Watch out for: Insurance does not replace security. Follow the risk alerts you receive.
How AI Agents Change Security Priorities
AI agents create risks most security programs were not built for.
- Shadow agents: Employees connect AI tools without IT approval.
- Overbroad access: Agents often get more permissions than they need.
- Prompt injection: Hidden instructions in documents can trick agents into harmful actions.
- Weak audit trails: Many teams cannot see what agents did or why.
Tools like Okta for AI Agents and Material’s OAuth Remediation Agent target these gaps directly.
How to Choose an AI Cybersecurity Vendor
Use this checklist:
- Start with your biggest risk. Email fraud, identity, code, or fraud each need different tools.
- Test on real data. Run a proof of concept in your own environment.
- Measure false positives. Noisy tools exhaust your team.
- Check automation limits. Know what the AI can do without human approval.
- Confirm integrations. Make sure it works with your identity, email, and cloud stack.
What Doesn’t Work in AI Security
AI helps, but it cannot fix weak basics.
- Skipping fundamentals: Multi-factor login, patching, and backups still matter most.
- Automating responses too early: Let AI recommend first. Automate once accuracy is proven.
- Ignoring people: Staff training stops many attacks that tools miss.
- Buying overlapping tools: Too many products create noise and gaps.
A real example: a fintech’s finance team receives a deepfake video call from a “CFO” asking for an urgent wire. A policy requiring callback verification on a known number stops the fraud. Technology plus process beat the attack.
Final Thoughts
AI cybersecurity companies in San Francisco protect every major attack surface. Abnormal AI and Material Security defend email and workspaces. Okta and Persona secure identities. Exaforce, Semgrep, and Socket strengthen operations and code. Doppel, Sardine, and Coalition fight impersonation, fraud, and financial risk.
Start with your top threat, test vendors on your own data, and expand automation only as trust grows.
FAQs
What are the top AI cybersecurity companies in San Francisco?
Leading names include Abnormal AI, Material Security, Okta, Persona, Exaforce, Semgrep, Socket, Doppel, Sardine, and Coalition.
How does AI detect phishing emails?
AI learns normal communication patterns. It flags messages that break them, even when they contain no malicious links.
How do you secure AI agents?
Give agents their own identities, limit their access, monitor actions, and be able to revoke access instantly.
Can AI detect deepfakes?
Yes. Identity and fraud tools use AI to detect spoofed faces and voices, though no detection method is perfect.
Does cyber insurance replace security tools?
No. Insurance covers financial losses. You still need security tools and good practices to prevent attacks.
