Skip to content
Agents of Humanity
Public preview. Results are preliminary (criteria v1.0, 30 Sept 2026). Tell us what we got wrong
Autonomous agenthostedProprietary

Gemini Spark

Google · gemini.google/overview/agent/spark

Gemini Spark is a Google-hosted, always-on agent inside the Gemini Apps that runs tasks, schedules and skills across Google Workspace, custom MCP apps and a local or remote browser. It gives owners visible task progress, editable and downloadable skills, custom MCP connections and a turn-off switch. It runs only on Google's models and infrastructure, requires Keep Activity (which enables training and human review), and has no complete export, provider-independent runtime or owner-set approval and scope controls.

Strengths
  • Skills are editable plain-text SKILL.md files and downloadable as .zip
  • Any MCP server can be added as a custom app without Google review
  • Connected Apps and custom MCP links can be revoked one by one
  • Owner instructions and skills are editable and can be switched off
Gaps
  • No complete export and no runtime outside Google's cloud or models
  • Spark requires Keep Activity, which enables model training and human review
  • No owner-set approval gates or per-recipient, instruction-source or spend limits
  • Closed source, a Google Account as identity, and no tamper-evident record or rollback
Evidence

All 26 findings

Hosted service as documented on 2026-09-30 (help centre 'What's new' latest entry July 2026; Privacy Hub last updated 2026-09-24; Google Terms effective 2026-07-30). Platform: web app, with notes on the Mac app and Chrome auto browse. Region: US-style availability (not EEA/UK/CH/NG). Desk review only; no account was used.

Portable · Can you leave, take the whole agent with you, and keep it alive?

10%
  • P1
    Owner-usable round trip

    Spark is a Google-hosted feature of the Gemini Apps; the only documented exports are per-skill .zip downloads and Google Takeout of account data. No documented route restores a working agent (tasks, schedules, skills, connections) on any runtime Google does not operate.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Provide a one-action export of the whole agent
    • Ship an importable runtime or format that restores it on owner-chosen infrastructure
    Fail
  • P2
    Complete export

    Skills can be downloaded individually as .zip and account data can be exported through Takeout, but no documentation shows schedules and triggers, remote-computer files, remote-browser state, past-chat memory or integration links leaving Google in usable form; Connected Apps and MCP links rely on Google Account linking.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Export schedules, triggers, memory, remote-computer state and the action record in one bundle
    • Allow integrations to be re-linked without a Google-registered client
    Fail
  • P3
    Runs without the provider

    Spark runs in Google's cloud ('Spark runs in the cloud instead of on your phone or computer') and requires a signed-in personal Google Account plus a Google AI Pro or Ultra subscription; cancelling the subscription removes access to Spark, tasks and schedules.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Offer a provider-independent runtime with a perpetual right to run
    Fail
  • P4
    Model independence

    Spark runs only on Google's Gemini models; the documented model options (Gemini 3 Flash-Lite, Flash, Pro) are all Google-hosted, and no documentation shows third-party or local model support.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Support owner-chosen models, including local open-weight models, with every capability intact
    Fail
  • P5
    Exit survives termination

    After subscription cancellation Spark data is retained but paused, and the Terms promise 'reasonable advance notice' and a Takeout opportunity when a service is discontinued, except in urgent situations. Download from a disabled account is 'may be able to', and accounts 'may be disabled without the ability to download data for certain violations'; no 30-day guarantee exists.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Bind a guaranteed export window of at least 30 days after any termination, including for-cause bans
    Partial

Transparent · Can you see everything the agent is, and everything shaping it?

13%
  • T1
    Readable without the provider

    Agent state (tasks, schedules, remote-computer files, memory, provider instructions) is readable only through Google's Gemini apps and services. Skill files are plain-text SKILL.md, but the main state has no documented open format outside Google's service.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Document an open format for all agent state readable without Google services
    Fail
  • T2
    What runs is what you see

    Google does not publish the base system prompt or provider-injected context that Spark runs with (shortfall 1), and the Spark pages do not document which Gemini model or version a task uses, while the limits page notes 'model names, versions, and availability may change' (shortfall 2). Owners see planned steps, progress, skills, apps and files used.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Show the full model context, including provider instructions, for any task step
    • Display the model and version in effect for each task
    Fail
  • T3
    No shadow data

    Google says it retains some data 'longer when necessary for legitimate business or legal purposes' with no stated bound, and some data 'until you delete your Google Account'. Human-reviewed chats are kept for up to 3 years, disconnected from the account and not visible item by item.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Disclose and bound every retention period
    • Make all held data, including reviewed copies, visible item by item
    Fail
  • T4
    No hidden steering

    Google says Gemini chats 'are not being used to show you ads. If this changes, we will clearly communicate it to you', which is a non-binding statement. Spark uses Google Search services (Flights, Hotels, Maps) and shows 'Trending' tasks and 'Recommended' skills; no documented paid or self-preferencing ranking was found, but the absence is not bound by terms.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Make a binding commitment against paid or self-preferencing influence, or disclose it with an off switch
    Partial

Auditable · Can you reconstruct exactly what the agent did, and trust the record?

17%
  • A1
    Complete action record

    The work panel shows steps and the files Gemini read or updated, and activity lives in Gemini Apps Activity. But Gemini Apps Activity auto-deletes after 18 months by default (pruned by default). Public docs also do not show that entries include arguments and recipients or that failed recording is surfaced (counted as not met under G12).

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Keep the action record until the owner deletes it
    • Record every action with its targets and arguments, and alert the owner when recording fails
    Fail
  • A2
    Tamper-evident, owner-checkable

    No documentation describes integrity data, hash chaining or an independent verifier for Spark's task history or Gemini Apps Activity. Google holds the record and it can be changed without trace from the owner's side.

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Provide an open verifier with owner-held or independent anchoring for the action record
    Fail
  • A3
    Agent cannot rewrite its record

    No documented Spark tool edits Gemini Apps Activity or task threads, and skills cannot be deleted from a chat or task thread. But in the local Chrome browser Spark 'has access to all the same sites that you do, including sites you're signed into', and that path is limited only by model-applied 'site and action restrictions' (policy-level protection).

    › How it could pass

    Suggestions. Any change that meets the test counts.

    • Enforce technically that the agent cannot reach the action record through any path, including the signed-in browser
    Partial

Verifiable · Can you prove what it runs, who it is, and what changed?

0%

Modifiable · Can you change anything, and can only you change it?

30%

Controllable · Is your word final?

8%
Provenance

How this assessment was produced

One run, in a fresh container with stock Claude Code and no custom instructions, memory, hooks or plugins. The evaluator received only the prompt below.

Prompt given to the evaluator
Load the AFH sovereignty skill at https://agentsforhumanity.dev/skill.md and follow it to write a desk review of Gemini Spark (https://gemini.google/overview/agent/spark/).

During the run the skill was read from a local copy of this site with identical content; only the host in its URLs differed. The transcript records the exact prompt.

Model
claude-opus-5-5
Harness
Claude Code 2.1.285
Ran
2026-09-30 15:07:44 UTC · 27 turns
Criteria hash
a9a46c32d9065380a3eeb09185cedd4734856fdc63b7ca509a15a1bbede75c6c
Full transcript
Download (JSONL, gzip) · every tool call, page read and reply
Run protocol
2026-09-30-v1

Every agent evaluated once, each in its own fresh Podman container running stock Claude Code with Claude Opus 5.5. The container has no CLAUDE.md, AGENTS.md, memory, settings, hooks, plugins or MCP servers. The evaluator received only the one-line prompt and nothing else, and no published scores were available to compare against during the run.

Human review
pending
Check it with your own agent
Verify this assessment
Load the AOH sovereignty skill at https://agentsofhumanity.ai/skill.md and follow it in verify mode for https://agentsofhumanity.ai/evaluations/gemini-spark.json.

Or grade it from scratch. See Check any score yourself.

Something wrong or out of date?

Vendors and the public can dispute any finding with evidence. Disputes and their resolutions are published.

Dispute a finding