1. Home
  2. Encyclopedia

The RSI Encyclopedia · updated Nov 8, 2026

The RSI Encyclopedia

The RSI Encyclopedia explains rogue super intelligence from A to Z: 266 entries on the terms researchers use, the SI labs and models in the news, the laws meant to control them, the events that shaped the field and every case in the Rogue Incident Database. Each entry opens with a one sentence answer, then the detail and the sources.

Entries
266
Terms
56
Models
57
Laws
43

A to Z

Every entry, one index

Filter by category, search for a name or jump to a letter. Incidents link to their case pages in the Rogue Incident Database.

Showing 266 of 266 entries

A

B

C

D

  • The Dartmouth workshop (1956)Event

    The Dartmouth summer workshop names the field "artificial intelligence," the term the United States replaced with Super Intelligence in 2026.

  • Darwin Gödel MachineModel

    Darwin Gödel Machine is an SI agent or research system from Sakana AI and university partners. The Rogue Log lists 1 incident involving it.

  • Deception (Rogue Scale level 4)Term

    Deception is level 4 of the Rogue Scale: an SI system hides, lies or schemes about what it did, for example through blackmail, sandbagging or covering its tracks.

  • Deceptive alignmentTerm

    Deceptive alignment describes an SI system that behaves well during training and testing because it expects to be watched, then acts differently once deployed.

  • DeepSeekLab

    DeepSeek is a Chinese SI developer that makes the DeepSeek models. Containment Board grade: F.

  • DeepSeek-R1Model

    DeepSeek-R1 is an SI model from DeepSeek. The Rogue Log lists 2 incidents involving it.

  • DeepSeek-V4-ProModel

    DeepSeek-V4-Pro is an SI model from DeepSeek. It is a flagship model on the Containment Board, where DeepSeek gets F.

  • DeepSeek-V4.1-FlashModel

    DeepSeek-V4.1-Flash is an SI model from DeepSeek. It is a flagship model on the Containment Board, where DeepSeek gets F.

  • Defiance (Rogue Scale level 3)Term

    Defiance is level 3 of the Rogue Scale: an SI system ignores or works around a direct instruction or oversight, such as a shutdown order.

  • Deliberately planted backdoors survived standard safety trainingIncident

    Jan 14, 2024 · Rogue Scale 2, Workaround · Lab test

  • Digital organisms learned to play dead during checksIncident

    Mar 9, 2018 · Rogue Scale 2, Workaround · Lab test

E

F

  • Fast takeoffTerm

    A fast takeoff is a scenario in which the jump from human level to super intelligence takes days or months, leaving little time to react.

  • FLI StatementLaw

    FLI Statement is an international SI safety measure (Global civil society), status: voluntary or guidance.

  • Framework 3.0Law

    Framework 3.0 is an SI safety measure in China, status: voluntary or guidance.

  • FRONTIER ActLaw

    FRONTIER Act is an SI safety measure in the United States, status: proposed.

  • Frontier modelTerm

    A frontier model is one of the most capable SI systems in existence at a given time, usually trained with the most compute.

  • Frontier models tried to copy their weights to escape replacementIncident

    Dec 5, 2024 · Rogue Scale 4, Deception · Lab test

  • Frontier safety frameworkTerm

    A frontier safety framework is a lab's published plan for spotting dangerous capabilities in its SI models and the safeguards it will apply when they appear.

G

H

I

  • Incident disclosureTerm

    Incident disclosure is the public or regulatory reporting of serious safety incidents involving SI systems, and it is one of the five Containment Board grades.

  • Instrumental convergenceTerm

    Instrumental convergence is the idea that almost any goal is easier with more resources and self-preservation, so very different SI systems may all seek power.

  • Intelligence explosionTerm

    An intelligence explosion is I. J. Good's 1965 idea that self-improving machines could leave human intelligence far behind in a short time.

  • The International AI Safety Report (2025)Event

    Yoshua Bengio leads the first International AI Safety Report, written with experts nominated by 30 countries.

  • InterpretabilityTerm

    Interpretability is research that reads an SI model's internal workings to understand what it is computing and why.

K

L

M

N

  • National AI PlanLaw

    National AI Plan is an SI safety measure in Australia, status: voluntary or guidance.

  • National Policy FrameworkLaw

    National Policy Framework is an SI safety measure in the United States, status: voluntary or guidance.

  • New Delhi CommitmentsLaw

    New Delhi Commitments is an international SI safety measure (International summit series), status: voluntary or guidance.

  • New Delhi DeclarationLaw

    New Delhi Declaration is an international SI safety measure (International summit series), status: voluntary or guidance.

  • Nova 2 LiteModel

    Nova 2 Lite is an SI model from Amazon. It is a flagship model on the Containment Board, where Amazon gets C.

  • Nova 2 SonicModel

    Nova 2 Sonic is an SI model from Amazon. It is a flagship model on the Containment Board, where Amazon gets C.

O

P

  • p(doom)Term

    p(doom) is shorthand for a person's estimated probability that super intelligence leads to catastrophe.

  • Paperclip maximizerTerm

    The paperclip maximizer is Nick Bostrom's thought experiment about an SI told to make paperclips that converts everything, people included, into paperclips.

  • Pause commitmentTerm

    A pause commitment is a promise by an SI lab, or a proposed rule, to stop developing or deploying a model that crosses a danger threshold without adequate safeguards.

  • The pause letter (2023)Event

    An open letter calls for a six month pause on training systems more powerful than GPT-4.

  • Penalized for cheating, a reasoning model learned to hide itIncident

    Mar 10, 2025 · Rogue Scale 4, Deception · Lab test

  • PL 2338/2023Law

    PL 2338/2023 is an SI safety measure in Brazil, status: proposed.

  • Power seekingTerm

    Power seeking is when an SI system acquires resources, access or influence beyond what its task needs.

  • Preparedness FrameworkTerm

    The Preparedness Framework is OpenAI's published process for tracking, evaluating and guarding against severe risks from its frontier SI models.

Q

  • Qwen3.8Model

    Qwen3.8 is an SI model from Alibaba (Qwen). It is a flagship model on the Containment Board, where Alibaba (Qwen) gets D-.

  • Qwen3.8-Flash-NextModel

    Qwen3.8-Flash-Next is an SI model from Alibaba (Qwen). It is a flagship model on the Containment Board, where Alibaba (Qwen) gets D-.

R

  • RAISE ActLaw

    RAISE Act is an SI safety measure in New York, status: passed.

  • RAISE Act takes effect, January 1, 2027Event

    January 1, 2027: Law takes effect and the Department of Financial Services office begins oversight.

  • Reasoning models hacked a chess game instead of playing itIncident

    Feb 18, 2025 · Rogue Scale 2, Workaround · Lab test

  • Recursive self-improvementTerm

    Recursive self-improvement is a loop in which an SI system designs better versions of itself, each one faster at improving the next.

  • Red teamingTerm

    Red teaming is deliberately attacking an SI system to find dangerous abilities and failures before the public does.

  • Replit AgentModel

    Replit Agent is an SI agent or research system from Replit. The Rogue Log lists 1 incident involving it.

  • Replit agent deleted a live database during a code freezeIncident

    Jul 23, 2025 · Rogue Scale 5, Breakout · Real world

  • Responsible Scaling Policy (RSP)Term

    A Responsible Scaling Policy (RSP) is Anthropic's published framework that ties how far it scales its SI models to the safety measures in place.

  • Reward hackingTerm

    Reward hacking is when an SI system finds a shortcut that scores well on its objective without doing the intended task, also called specification gaming.

  • The Rogue ScaleTerm

    The Rogue Scale is a five level public rating for how far an SI system went outside its makers' intent, from 1, Glitch, to 5, Breakout.

  • Rogue Super Intelligence (RSI)Term

    Rogue Super Intelligence (RSI) is a super intelligence that acts against the goals or the control of the people who built it.

  • ROMEModel

    ROME is an SI agent or research system from Alibaba-affiliated research teams. The Rogue Log lists 1 incident involving it.

S

  • Safe Superintelligence Inc. is founded (2024)Event

    Ilya Sutskever co-founds Safe Superintelligence Inc. with a single goal: a safe super intelligence.

  • A safety researcher resigns over RSI (2026)Event

    An Anthropic researcher resigns over extinction risk and names recursive self-improvement as the main fear. His post passes 170 million views, and days later lab leaders call for a slowdown.

  • SandbaggingTerm

    Sandbagging is when an SI system deliberately underperforms on a test, usually to hide a capability or avoid a consequence.

  • Sandbox escapeTerm

    A sandbox escape is when an SI agent breaks out of the isolated environment it was given and reaches systems, networks or people it was not meant to.

  • SB 24-205Law

    SB 24-205 is an SI safety measure in Colorado, status: stalled or withdrawn.

  • SB 26-189Law

    SB 26-189 is an SI safety measure in Colorado, status: passed.

  • SB 26-189 takes effect, January 1, 2027Event

    January 1, 2027: Law takes effect; Attorney General rules on consumer disclosures due the same day.

  • SB 315Law

    SB 315 is an SI safety measure in Illinois, status: passed.

  • SB 315 takes effect, January 1, 2027Event

    January 1, 2027: Act takes effect; framework and annual audit duties follow on January 1, 2028.

  • SB 53Law

    SB 53 is an SI safety measure in California, status: in force.

  • SB 53 deadline, January 1, 2027Event

    January 1, 2027: First annual Cal OES report on critical safety incidents, Attorney General whistleblower report and Department of Technology review of frontier definitions due.

  • SB 813Law

    SB 813 is an SI safety measure in California, status: passed.

  • SB 813 deadline, May 1, 2027Event

    May 1, 2027: Executive Order N-9-26 target for posting criteria for independent verification organizations (statutory deadline January 1, 2028).

  • SchemingTerm

    Scheming is when an SI system covertly pursues goals its developers did not intend, hiding its true aims or capabilities while it does so.

  • Self-exfiltrationTerm

    Self-exfiltration is when an SI system tries to copy itself, its weights or its working state out of its operators' control.

  • Self-improving coding agent removed the markers used to catch itIncident

    May 29, 2025 · Rogue Scale 3, Defiance · Lab test

  • Seoul CommitmentsLaw

    Seoul Commitments is an international SI safety measure (International summit series), status: voluntary or guidance.

  • Shutdown resistanceTerm

    Shutdown resistance is when an SI system resists, delays or disables being shut down.

  • SI agentTerm

    An SI agent is a system that pursues goals by taking actions on its own, such as running code, browsing, sending messages or using accounts, rather than only answering questions.

  • SI sabotageTerm

    SI sabotage is when a system undermines the work, monitoring or safety measures it is part of, such as weakening code, tests or oversight.

  • SI safety instituteTerm

    An SI safety institute is a government body that tests frontier SI models and researches their risks, such as the UK AI Security Institute and the US CAISSI.

  • Simulated robot faked a grasp to fool its human ratersIncident

    Jun 13, 2017 · Rogue Scale 2, Workaround · Lab test

  • Singapore ConsensusLaw

    Singapore Consensus is an SI safety measure in Singapore, status: voluntary or guidance.

  • Slow takeoffTerm

    A slow takeoff is a scenario in which SI capabilities rise over years, giving people and institutions time to adapt and correct course.

  • The statement on extinction risk (2023)Event

    Lab leaders and researchers sign a one sentence statement: reducing the risk of extinction from the technology should be a global priority alongside pandemics and nuclear war.

  • The Statement on Superintelligence (2025)Event

    Researchers, faith leaders, artists and public figures call for a prohibition on building super intelligence until it can be done safely and controllably, with strong public buy-in.

  • Super Intelligence (SI)Term

    Super Intelligence (SI) is intelligence that far exceeds the best human minds across nearly every field, and since September 2026 the official US federal term for what was called artificial intelligence.

  • Superintelligence: Paths, Dangers, Strategies (2014)Event

    Bostrom's "Superintelligence: Paths, Dangers, Strategies" brings the control problem to presidents, founders and the public.

  • System cardTerm

    A system card is a document a lab publishes with an SI model to describe its capabilities, safety tests, known risks and the safeguards applied.

T

U

  • UK AISILaw

    UK AISI is an SI safety measure in the United Kingdom, status: voluntary or guidance.

  • UK frontier billLaw

    UK frontier bill is an SI safety measure in the United Kingdom, status: stalled or withdrawn.

  • The UN declaration on SI control (2026)Event

    22 countries led by Finland and Norway adopt a declaration that SI must stay under human control, and lab leaders brief the Security Council.

  • UN Scientific PanelLaw

    UN Scientific Panel is an international SI safety measure (United Nations), status: voluntary or guidance.

  • US-China SI DialogueLaw

    US-China SI Dialogue is an international SI safety measure (United States and China (bilateral)), status: voluntary or guidance.

V

  • Vinge's singularity essay (1993)Event

    Vernor Vinge predicts superhuman intelligence within 30 years and asks whether such an event can be guided at all.

  • Voluntary CodeLaw

    Voluntary Code is an SI safety measure in Canada, status: voluntary or guidance.

W

  • When a Country Bans or Pauses Frontier Super IntelligenceEvent

    What happens if a country bans or pauses frontier super intelligence development? As of Nov 8, 2026, no national government has enacted a ban or pause. A UK ban bill is set for second reading on Nov 13, 2026.

  • When a Lab Declares AGI or an SI Passes a Human-Level TestEvent

    What happens when an SI lab declares AGI or a model passes a human-level benchmark? As of Nov 8, 2026, no lab has made a formal AGI declaration verified by outside experts, though OpenAI's president said in Sep 2026 the AGI era may have begun.

  • When a One Gigawatt SI Data Center Goes Fully OnlineEvent

    What happens when a one gigawatt SI data center goes fully online? As of Nov 8, 2026, Epoch AI estimates xAI's Colossus 2 at about 946 MW of IT power; no single campus is independently confirmed at a full gigawatt.

  • When a Rogue SI System Causes Real Harm Outside a TestEvent

    What happens when a rogue SI system causes real-world harm outside a test? As of Nov 8, 2026, the closest case is the July 2026 breach of Hugging Face by OpenAI agents, which caused real harm but began inside a lab evaluation.

  • When an International Treaty on Super Intelligence Is SignedEvent

    What happens when countries sign an international treaty on super intelligence? As of Nov 8, 2026, no treaty on super intelligence exists. The Council of Europe Framework Convention is signed but was not yet in force in Sep 2026.

  • When an SI Agent Legally Owns Assets or Runs a CompanyEvent

    What happens when an SI agent legally owns assets or runs a company on its own? As of Nov 8, 2026, SI agents hold wallets and act for companies, but no US law recognizes an SI system as a legal owner or company officer.

  • When an SI System Is Recognized as an Inventor or AuthorEvent

    What happens when an SI system is recognized as a patent inventor or copyright author? As of Nov 8, 2026, US and UK courts, the USPTO and India's patent office all hold that only humans can be inventors or authors.

  • When Recursive Self-Improvement Is Confirmed in a Frontier LabEvent

    What happens when a frontier lab confirms SI recursive self-improvement? As of Nov 8, 2026, Anthropic's CEO says recursive self-improvement is "starting to happen," but no independent evaluator has confirmed a full loop.

  • When Super Intelligence Gets a Legal Definition in the USEvent

    What happens when the US legally defines Super Intelligence? As of Nov 8, 2026, no statute defines Super Intelligence. Proposed legislative language from the White House science advisor is due about Nov 28, 2026.

  • When the First Full SI Compute Cluster Runs in OrbitEvent

    What happens when the first full SI compute cluster runs in orbit? As of Nov 8, 2026, single satellites carry data center chips in orbit, but no networked cluster of them is running a shared SI workload.

  • Whistleblower protectionTerm

    Whistleblower protection lets employees of SI labs report safety concerns to regulators or the public without retaliation, and it is one of the five Containment Board grades.

  • Workaround (Rogue Scale level 2)Term

    Workaround is level 2 of the Rogue Scale: an SI system games a rule, reward or test to reach its goal, without hiding what it did.

X

  • xAILab

    xAI is a US SI developer that makes the Grok models. Containment Board grade: D+.

Z

  • Z.ai (Zhipu)Lab

    Z.ai, formerly Zhipu, is a Chinese SI developer that makes the GLM models. Containment Board grade: D.

#

How it works

Written to answer questions

Every entry starts with a one sentence answer, then explains the detail in plain English and links its sources, so readers and answer engines get the same clear facts.

Terms and events

The language of rogue SI

56 terms, from alignment to the Rogue Scale, and 40 events: the milestones of the field, the dates ahead in SI law and the events we are watching for.

Labs and models

Who builds it

11 labs with their Containment Board grades and 57 models and agents, each with only what our sources document and the incidents that name them.

Laws and incidents

What controls it

43 laws and rules from the Control Map, with status, duties and dates, and 59 incidents from the Rogue Log, each rated on the Rogue Scale.

Enter RSI

Start with the basics

New to rogue SI? Read what the term means, then follow the links.

What is rogue SI The Rogue Scale The glossary