Is Granola Safe? Security, AI Training, and Settings to Use It Safely [October 2026]

Hi, Hoda here. In this article, I’ll go over Granola.ai’s security, based on third-party analyses, and sort out what’s reassuring and what you should watch out for.

I’m sure many of you are wondering, “Is Granola safe?”

It’s a tool that entrusts your meeting content to AI, so the concern is only natural. To get to the conclusion: Granola is a tool you can use with confidence for everyday business meetings. However, I don’t recommend using it with the default settings.

I use Granola myself for client meetings and sales calls. I recommend it because I understand it’s safe, but I’ll also cover everything you should be careful about, so you don’t end up using it without knowing and running into trouble later.

This article draws on Granola’s official security page, help center, and privacy policy, along with third-party analyses, to lay out what’s reassuring and what needs attention.

The Bottom Line: Granola Is Safe to Use, but Review Your Settings First

Here’s the conclusion up front.

While the way the security is built can be trusted, you need to manage the default settings and consent or notice to other participants yourself. That’s the main message of this article.

How Granola Works: Why “No Bot” Matters for the Safety Discussion

Granola doesn’t use a bot that joins your meetings. It captures the audio from your own computer (or phone) microphone along with the meeting audio coming out of your computer, transcribes it in real time, and has AI shape it into notes alongside your own notes.

• You start meetings manually (it never joins or records automatically)
• Works with any meeting tool, including Zoom, Google Meet, and Teams
• Audio is processed temporarily just for transcription, then deleted
• Only the transcript text and notes are stored

For the person using it, having no visible bot is convenient and doesn’t disrupt the atmosphere of the meeting. On the other hand, it also means other participants may not easily notice they’re being recorded, which leads to the consent and lawsuit issues discussed below.

I introduce how Granola works and how to use it in the video below!

What’s Reassuring

SOC 2 Type 2 Certification

Granola has obtained SOC 2 Type 2 (July 2025). While Type 1 looks at whether controls are in place at a single point in time, Type 2 has a third party audit whether those controls actually operate over a period of 6 to 12 months. It’s one of the first things checked in a company’s IT security review.

You can request access to the report itself from Granola’s Trust Center.

Encryption and Data Storage

• Notes and transcripts are encrypted at rest and in transit on AWS (US)
• Daily backups
• The local cache on your device (for offline viewing) has been encrypted since 2026

Audio Isn’t Stored

Granola doesn’t store audio. The desktop app (Mac/Windows) transcribes in real time, while the mobile app temporarily caches audio for post-meeting transcription. In both cases, the audio is deleted after processing. You can’t play back the recording afterward, which can be inconvenient, but it reduces the risk of audio data being leaked.

External AI Providers Can’t Train on Your Data

AI providers such as OpenAI and Anthropic, which are used for summarization and other features, are not permitted to use your data to train their models.

GDPR and UK GDPR Compliance

A standard Data Processing Agreement (DPA) is available, and it incorporates the EU and UK Standard Contractual Clauses (SCCs). You can also delete individual notes, and account deletion is supported (data is deleted within 30 days).

Past Vulnerabilities Were Disclosed and Fixed

Granola has published postmortems for past security issues (all in 2025), such as an exposed transcription API key in the iOS beta, a Google account session-management issue, and a flaw in workspace auto-join. There were problems, but the fact that they were disclosed openly and fixed is worth recognizing. No new incidents have been confirmed in 2026 (as of October 2026).

In addition, security firm PromptArmor’s analysis (April 2026) found that the desktop app includes protections against data leakage and phishing through AI output (disabling Markdown images, warnings for external links, and disabling HTML). The flaws found in the mobile app were also addressed in late March 2026 after being reported.

What to Watch Out For

This is the part you should know before you start using it.

1. AI Training Is On by Default (Free and Business)

On the Free and Business plans, the setting that lets Granola improve its own AI models with anonymized data is ON by default. You can turn it OFF yourself in settings, but there are three things to keep in mind.

• Opting out (turning it OFF) applies only to data going forward. There’s no guarantee that earlier data wasn’t used for training
• On the Business plan, each user has to turn it OFF individually (admins can’t turn it off for everyone at once)
• The Android app doesn’t have this toggle yet (changing it on desktop or iOS applies across the board)

On the Enterprise plan, training is OFF by default for the whole organization.

What to do: Turn this setting OFF before your first meeting.

Settings → Preferences → Data & sharing → Turn OFF “Use my data to improve models for everyone”

2. Recording Consent Is Your Own Responsibility

This is the most important issue with Granola.

Because no bot joins, participants may not realize their speech is being transcribed. And Granola itself states that obtaining consent from participants is the user’s responsibility.

Granola has two notification features.

• Automatic chat message: Posts a message like “Granola is transcribing this call” to the meeting chat
• Watermark: Shows an indicator on your own video feed that transcription is in progress

However, neither is active until you turn it on yourself (on Enterprise, admins can enforce it across the organization).

Be especially careful when speaking with people in places that require consent from everyone in the conversation (all-party consent), such as California in the US. Even where consent isn’t legally required, getting into the habit of telling the other person is important for building trust.

What to do: Say something like, “I’m taking notes with AI. Is that okay with you?” Also turn on the chat message and the watermark.

Settings → Preferences → Transparency

3. Notes Are Stored Indefinitely

Notes and transcripts are stored indefinitely unless the user or an admin sets a retention period. The longer data is kept, the greater the impact of an accidental leak, and the greater the chance it gets viewed unintentionally within your company.

• Automatic transcript deletion can be configured on all plans (1 day to 1 year). However, only the transcript is deleted; the notes themselves remain
• There’s a one-week cooldown when you change personal settings
• On Enterprise, admins can set an organization-wide deletion policy

What to do: If you have many sensitive meetings, set automatic transcript deletion to a short period, and manually delete the notes themselves when you no longer need them.

4. Data Is Stored Only in the US

Data is stored in AWS’s US region. You can’t choose an EU, UK, or other region. SCCs (Standard Contractual Clauses) apply to cross-border data transfers, but it isn’t suited to organizations that require data to be kept in their own country or region.

5. The Class Action Lawsuit (Chamberlain v. Granola)

On July 30, 2026, a class action lawsuit against Granola, Inc. (Chamberlain v. Granola, Inc.) was filed in the US District Court for the Northern District of California.

The complaint broadly makes two allegations.

• Granola’s bot-free recording captures conversations without the consent of the other participants
• It uses the captured content to train AI models by default

The important point is that these are the plaintiff’s allegations and nothing has been established so far. As of October 10, 2026, the court record shows no answer or motion to dismiss from Granola. Federal wiretapping law generally follows “one-party consent,” which allows recording if one party to the conversation consents, so the plaintiffs face a high bar (though state laws can be stricter).

Similar lawsuits have also been filed against other AI meeting-notes tools such as Otter.ai and Fireflies.ai, so this is an issue the whole industry is facing, not something unique to Granola.

My practical takeaway from this lawsuit comes down to the two points above:

• Check your training setting (turn it OFF)
• Tell people you’re recording

If you do both of these yourself, you can greatly reduce your own risk related to the issues in the lawsuit. The situation is evolving, so check for the latest updates if you’re interested.

6. Link-Sharing Settings

Notes are private by default, and even admins can’t see other people’s private notes. However, you need to be careful with how links are shared.

• The default link-sharing option can be set to “Anyone with the link,” “Only my company,” or “Private”
• The official documentation doesn’t specify which is the initial default, so check your own setting and set it to “Only my company” or “Private” to be safe
• A new default applies only to notes created after you change the setting. Check any links you’ve already shared separately
• Someone who opens a link in a browser sees only the summary notes, while collaborators in the desktop app can also see the full transcript

Also, PromptArmor’s analysis pointed out that if a logged-out viewer asks questions in chat on a shared note, their input may fall outside the data protections in Granola’s contract. If your company shares notes externally, consider reviewing the contract terms and limiting the sharing scope.

7. Other Smaller Points

• No two-factor authentication (2FA) as standard: Authentication is delegated to your identity provider. SSO is available only on Enterprise with 50 or more seats. Enable two-step verification on the account you use to log in, such as your Google account.
• Chat model on the Business plan: According to Granola’s documentation, the selectable Anthropic model (Fable 5) isn’t covered by zero data retention, and Anthropic keeps logs for 30 days. For chats about sensitive meetings, it’s safer to choose a different model.
• Scope of certifications and compliance: ISO 27001 hasn’t been obtained, and FERPA (US education records) isn’t supported.
• Integrations: When you send notes to other services through Slack, Notion, Zapier, MCP, and so on, the settings of those downstream services come into play. Limit integrations to the ones you need.

Security Differences by Plan

For individuals and small teams whose meetings aren’t highly sensitive, Basic or Business with a settings review is enough. If you handle medical information or want to apply policies across the whole organization, consider Enterprise.

When to Use Granola and When to Be Cautious

5 Steps to Use Granola Safely

Check the following before your first meeting. It takes under 10 minutes.

1. Turn AI training OFF: Settings → Preferences → Data & sharing → Turn OFF “Use my data to improve models for everyone” (on Business, everyone does this individually; on Android, change it from desktop or iOS)
2. Set automatic transcript deletion: Choose as short a period as is acceptable (1 day to 1 year). Note that the notes themselves aren’t deleted automatically
3. Turn on the chat message and watermark: Settings → Preferences → Transparency. Essential for meetings with people outside your organization
4. Review your default link-sharing setting: Settings → Preferences → Data & sharing → Set Default link sharing to “Only my company” or “Private.” Also check links you’ve already shared
5. Enable two-step verification on your login account: Granola doesn’t offer MFA as standard, so set it up on your Google or Microsoft account. On the Business plan, also pay attention to the chat model you use for sensitive meetings

In addition, make it a habit to say it out loud at the start of the meeting.

“I’m using Granola to transcribe for the meeting notes. Is that okay with you?”

Why I Recommend Granola

FAQ

Is Granola safe?

For everyday business meetings, yes, it’s a safe tool to use. It has safeguards including SOC 2 Type 2 certification, encryption at rest and in transit, no audio storage, and a ban on training by external AI providers. However, AI training is ON by default on Free and Business, so I recommend reviewing your settings.

Does Granola use my data to train AI?

External providers such as OpenAI and Anthropic aren’t permitted to use your data for training. Granola itself, on the other hand, uses anonymized data to improve its own models on the Free and Business plans (ON by default). You can turn it OFF in settings, but there’s no guarantee that data from before you turned it OFF wasn’t used. On Enterprise, it’s OFF by default.

Are recordings kept?

Audio isn’t kept. It’s deleted after transcription processing. What remains are the transcript text and notes, which are stored indefinitely by default.

Can it record without the other person knowing?

Yes. With the default settings, the other person can’t see that Granola is running. That’s why I recommend turning on the chat message and watermark and also telling people out loud. Responsibility for obtaining participants’ consent lies with the user.

Is it okay to use in meetings outside the US?

Using it in ordinary meetings where you’ve notified participants is generally a low-risk way to use it. However, if the other party is in a place that requires all-party consent (such as certain US states), get everyone’s consent. Rules differ by country and region, so consult a professional for legal judgments in your specific case.

Does it support HIPAA?

Only the Enterprise plan supports signing a BAA (Business Associate Agreement). It covers the desktop app only, not the mobile app. Don’t use the Free or Business plans for medical information.

Where is my data stored?

On AWS servers in the US. You can’t choose regions such as the EU or Asia.

Is it okay to keep using a tool that’s being sued?

The lawsuit is at the stage of the plaintiff’s allegations, and there’s no outcome yet. Similar lawsuits have been filed against other AI meeting-notes tools. As a user, the realistic approach is to follow the basics: turn training OFF and notify participants. The situation may change, so check for the latest information regularly.

Can I take my data out if I want to stop using Granola?

You can export notes and transcripts (CSV export is limited to once every 24 hours). If you delete your account, your data is deleted within 30 days.

Summary: Review Your Settings and Notify Participants, and You Can Use It with Confidence

• Granola’s technical security (SOC 2 Type 2, encryption, no audio storage) is solid
• The areas that need attention are default settings and how consent is handled
• Before you start, check these four things: training OFF, notifications ON, automatic deletion, and link sharing
• At the start of each meeting, tell people, “I’m transcribing”
• If you handle medical information or must keep data in your own country, Granola isn’t a good fit
• The class action lawsuit is pending with no conclusion yet. Check for the latest updates

With these in place, Granola becomes a strong ally for your daily meetings.

This article is based on official documentation and third-party analyses as of October 31, 2026. Specifications and the status of the lawsuit may change, so please check the official website for the latest information. If you need a legal judgment, consult a professional.

References:

• Granola Security page and Security/Privacy FAQ (official)
• Granola Help Center (official)
• Granola official blog (meeting safety checklist, guide for IT reviews)
• PromptArmor, “Granola AI Security Risks and Remediations” (April 2026)
• Voibe, “Is Granola Safe?” (October 2026), and other articles on Chamberlain v. Granola