The comparison
| ChatGPT (Free/Plus/Pro) | Ask Safely | |
|---|---|---|
| Trains on your data | Yes, by default. You can opt out in Settings > Data Controls, but the opt-out only applies to future conversations. Data already sent stays in the training set. | Never. Two layers: AWS Bedrock's contract prohibits training on customer data, and Ask Safely adds its own no-training policy on top. |
| How long chats are kept | Indefinitely, unless you delete them. Deleted chats are purged within 30 days. In 2025, a court ordered OpenAI to preserve deleted chats for months as part of the NYT copyright lawsuit. | Auto-deletes on a timer you choose: 8 hours (default), 7 days, or 30 days. A database TTL index handles deletion, not application code. When the timer hits zero, the conversation is gone. |
| Memory | Two layers: saved memories you can see and edit, plus a hidden "reference chat history" layer you can't fully audit. OpenAI's own FAQ says the visible summary "will not include everything that ChatGPT remembers." Deleting a chat does not delete memories made from it. In June 2026, OpenAI added "Dreaming V3," which reviews your history and updates memory on its own. | Off by default. Turned on only with your explicit consent. Every fact stored is visible, editable, and deletable. Declined facts leave no trace. Extraction stops the moment you turn it off. |
| Encryption | AES-256 at rest, TLS 1.2+ in transit. No per-user encryption keys on consumer tiers. | AES-256-GCM at rest with per-user data encryption keys, KMS envelope encryption. The memory system is fail-closed and refuses to run without its encryption key. |
| Advertising identifiers | No IDFA-based ad tracking. But the iOS app's privacy label declares data collection linked to your identity across Health & Fitness, Location, Contact Info, User Content, Search History, and Identifiers. It requests permissions for Camera, Location, and HealthKit. | No IDFA. No GAID. No ATT prompt. No advertising identifiers. No location, contacts, or photos. No email open/click tracking. |
| Human review | Not disclosed for consumer tiers in the same way Google discloses it. Feedback (thumbs up/down) may include the full conversation and can be reviewed. TaskUS listed as a human annotation subprocessor. | No human review of conversations. |
| Can your chats be subpoenaed? | Yes. ChatGPT conversations carry no legal privilege. Sam Altman said publicly: "if you go talk to ChatGPT about your most sensitive stuff and then there's a lawsuit, we could be required to produce that." In 2025, a court ordered OpenAI to produce 20 million de-identified chat logs. | Ask Safely can't produce what it doesn't keep. Chats auto-delete on the user's timer. After deletion, there is nothing to subpoena. |
| Price | Free / $8 (Go) / $20 (Plus) / $100-200 (Pro). Privacy features like default no-training and admin controls require Business ($20-25/seat/mo) or Enterprise (custom, ~$60/seat/mo). | Free (Essential) / $12/mo or $120/yr (Expert). Auto-delete and no-training are included on the free tier. |
What ChatGPT actually does with your data
OpenAI is not secretive about this. It's in their documentation. Most people just don't read it.
Training is on by default. Every conversation on a Free, Plus, or Pro account can be used to train future models unless you find the toggle and turn it off. The toggle is in Settings > Data Controls > "Improve the model for everyone." It's forward-looking only. Conversations you had before turning it off remain in the training pipeline.
"Temporary Chat" still keeps data for 30 days. Despite the name, OpenAI retains a copy for safety and abuse monitoring. Enterprise admins can retrieve Temporary Chat content via the Compliance API. If a Temporary Chat uses a third-party GPT with an external API, that data is governed by the third party's privacy policy, not OpenAI's.
Memory works in ways you can't fully see. ChatGPT stores two kinds of memory: explicit saved facts you can view and delete, and an implicit "reference chat history" layer that draws on patterns across your past conversations. OpenAI says the visible summary doesn't show everything it remembers. "Dreaming V3," rolled out in June 2026, reviews your chat history after each session and updates memory without asking. The Wall Street Journal reported cases of ChatGPT resurfacing sensitive past disclosures (a user's divorce, for example) in unrelated later conversations.
Deleted chats aren't necessarily deleted. OpenAI says deleted conversations are purged within 30 days. But in 2025, a federal court ordered OpenAI to preserve all chat logs, including deleted ones, as part of the NYT copyright case. For roughly four months, "deleted" meant "moved to a legal-hold archive." OpenAI says it returned to normal deletion after the order was lifted in September 2025, but still holds a slice of data from that period at the Times' demand.
No legal privilege. This is the one that matters most if you use ChatGPT for anything sensitive. Sam Altman said it plainly: conversations with ChatGPT have no legal confidentiality. They can be subpoenaed. The ABA's Formal Opinion 512 says lawyers must get informed client consent before putting case information into consumer AI tools. California courts have confirmed that AI chat logs don't qualify as privileged communication.
What Ask Safely does differently
Ask Safely runs on Anthropic's Claude models but wraps them in a different data architecture.
Chats auto-delete. You pick the timer at signup: 8 hours, 7 days, or 30 days. A MongoDB TTL index handles deletion at the database level. When the timer expires, the conversation is permanently removed. Swipe left to delete instantly.
No training. Period. Ask Safely accesses Claude through AWS Bedrock, which contractually prohibits using customer data for model training. Ask Safely adds a second layer: its own policy never sends data back for training. Two independent commitments, not one.
Memory is opt-in and transparent. Ask Safely's memory (called Knowledgebooks) defaults to off. When you turn it on, every fact is stored with attribution showing which chat it came from. You can read, edit, or delete any line. If you decline a suggested memory, it's hard-deleted with no trace. There's no hidden layer. No autonomous synthesis. No "the system remembers things you can't see."
Per-user encryption. AES-256-GCM with a per-user data encryption key wrapped by AWS KMS. The memory system is fail-closed: if the encryption key isn't available, the system refuses to operate rather than falling back to plaintext. This is the opposite of ChatGPT's infrastructure-level encryption, which doesn't isolate one user's data from another's at the cryptographic level.
Where ChatGPT is better
ChatGPT has features Ask Safely doesn't. Being honest about this makes the rest of the comparison worth trusting.
Plugin and GPT ecosystem. ChatGPT has thousands of third-party GPTs and integrations. Ask Safely has four guided agents (Researcher, Writer, Troubleshooter, Lookout) and an MCP connector coming in late October 2026.
Image generation. ChatGPT generates images with DALL-E. Ask Safely doesn't generate images.
SOC 2 and enterprise compliance. ChatGPT Enterprise has SOC 2 Type II, HIPAA BAAs, Enterprise Key Management, and SSO/SCIM. Ask Safely's SOC 2 is expected early 2027. If your employer or compliance team requires a signed BAA or SOC 2 report today, ChatGPT Enterprise (not the consumer tiers) meets that bar. Ask Safely doesn't yet.
Market presence. ChatGPT has hundreds of millions of users. Ask Safely has fewer than a thousand MAU. ChatGPT's scale means more community resources, more tutorials, and faster feature iteration.
The bottom line
If you don't care about privacy, ChatGPT is a great product. Use it.
If you care about privacy but not enough to pay or change habits, turn off "Improve the model for everyone" in ChatGPT's settings and use Temporary Chat for anything sensitive. It's imperfect (30-day retention, no legal privilege, memory you can't fully audit), but it's better than the default.
If you want an AI where the default is private and you don't have to find a toggle, change a setting, or wonder what it remembers about you, that's what Ask Safely is built for. Same Claude models. Different rules about what happens to your words after you say them.