Safenet Beta

Safenet Beta was the first live deployment of Safenet. It demonstrated that a decentralized network could evaluate Safe transactions, produce onchain attestations, and support SAFE staking and delegation. Safenet Beta and its reward program were approved by SafeDAO through SEP 55. The current Safenet architecture builds on Beta with clearer participant roles, a public Charter, stronger accountability, transaction fees, and optional onchain enforcement. SafeDAO also has greater control over protocol rules, participant sets, and configurable parameters. SAFE is used for Validator staking and delegation, participant incentives, and SafeDAO governance. Safenet Beta documentation and code remain available as historical references.

​

Beta in numbers

Safenet Beta demonstrated that the network could operate reliably at meaningful scale. TODO: Verify and source the final Beta metrics before publication.

• 6 Validators.
• ~98% Validator participation rate.
• ~900,000 transactions checked.
• ~800 SAFE stakers.
• 4.5 million SAFE distributed as rewards.

​

What changed after Beta

The largest architectural change is the separation between checking and attestation. In Safenet Beta, Validators evaluated transactions and produced attestations. In the current architecture:

• Sentinels evaluate transactions against the Safenet Charter.
• Validators attest secure Sentinel results.
• The Safenet Guard can enforce a valid secure attestation before normal execution.
• The Safenet Security Council handles arbitration when Sentinel decisions are split.

This separation gives each participant role a narrower responsibility and makes transaction evaluation independently attributable.

​

From five fixed checks to a Charter

Safenet Beta used five predefined transaction checks:

• No unexpected delegatecalls.
• Upgrade only to trusted Safe Singletons.
• Only allow adding trusted Safe Modules.
• Only trusted Safe Fallback Handlers can be set.
• Only trusted Safe Guards can be set.

These checks were static and fully deterministic. The current Safenet architecture uses the Safenet Charter, a public and versioned rulebook that Sentinels apply to each transaction. The Charter keeps deterministic Safe-configuration coverage, but extends the checking model to broader transaction-integrity risks, including manipulated recipients or targets, compromised addresses, delegatecall behavior, and approvals. This makes Safenet’s checking model more expressive and allows the rules to evolve through SafeDAO governance without changing the basic Sentinel and Validator roles. See Safenet checks for the current coverage.

​

Economics

Beta relied on SAFE-denominated subsidies to bootstrap Validator participation. The current Safenet model separates the economics of checking and attestation while expanding incentives for both participant roles:

• Sentinels receive USDC-denominated transaction-check fees and post per-check bonds. They also receive SAFE-denominated rewards under the current Sentinel reward program.
• Validators and Delegators participate through SAFE staking and receive SAFE-denominated rewards under the Validator and Delegator reward program.

Unlike Beta, where no slashing applied to any participant, Sentinel bonds can now be slashed under the applicable protocol rules. Validator stake remains unslashed. This introduces a direct fee-based incentive for transaction checking, while SAFE continues to support network participation and bootstrap both Sentinel and Validator incentives. See Economics for the current model.

​

Wallet integration

Safenet Aegis is designed to be integrated directly into wallet transaction flows. Wallet providers can integrate:

• Safenet checks: Show users the Safenet result before execution.
• Safenet Guard: Optionally require a valid secure attestation before normal Safe execution.

Safenet is permissionless to integrate. Compatible wallet providers can integrate Safenet without approval. Safe Wallet integration is planned as part of the Aegis rollout. See Wallet integration for the current integration model.