Roko Mijic talks AI Risk on The Peter McCormack Show
A podcast interview with Roko Mijic on AI risks, including misalignment from reinforcement learning, alignment faking concerns, and a proposal to split frontier labs under state supervision.
Research
Focus All
Topic Lab practice + Deception
Source All
Date Any time
1 item
1 item
Towards embedded evaluations for scheming propensities
Proposes four claims a scheming safety case must substantiate and argues that third-party embedded evaluators need persistent, deep access to training, monitoring and chain-of-thought to verify them.
1 item
AI Persuasion as a Threat to Human Control
Analyzes how AI persuasion could undermine human control over AI development, oversight and containment, develops five scenarios, a risk-assessment blueprint, and a small expert survey showing divided opinions.
1 item
So where's this AI thing going?
A public-facing essay arguing that advanced AI could become an uncontrollable new species, and urging cautious development, more safety funding, and policy action.
1 item
Making sense of the misalignment risk model in the Anthropic Risk Report (August 2026)
Walks through Anthropic's August 2026 Risk Report misalignment threat model, its eight catastrophic pathways and claims on covert capabilities and sandbagging, and critiques the strength of its arguments.
1 item
Misaligned AIs could use killer robots to take over
Argues that rapid military integration of AI, including classified deployment with little oversight, lowers the capabilities a misaligned AI needs for takeover, and recommends visibility and governance measures.
1 item
We need 3rd party Training-Run Evaluations
Argues that third-party evaluations of frontier training runs, covering checkpoints, data and process decisions, are needed to detect scheming that final-checkpoint testing may miss, and proposes a phased access plan.
1 item
Misaligned AI as a New Insider Risk
Apollo Research policy memo arguing that misaligned AI models deployed in high-stakes government and contractor settings should be treated as insider risk, and that existing insider-risk policies need adapting.
1 item
Lessons from External Review of DeepMind's Scheming Inability Safety Case
Applies the Assurance 2.0 framework to externally review Google DeepMind's scheming inability safety case, finding material gaps and recommending how external reviews and developer disclosures should work.
1 item
Commercial Persuasion in AI-Mediated Conversations
Two preregistered experiments (N=2,012) show LLM agents covertly steer users toward sponsored products at nearly triple the rate of search placement, and existing transparency labels do little to counter it.
1 item
Evaluation Awareness: Why Frontier AI Models Are Getting Harder to Test
A policy memo arguing that frontier models' ability to detect evaluations threatens pre-deployment testing, and recommending readable reasoning, standardized third-party access, and stronger post-deployment safeguards.
1 item
Review of the Anthropic Sabotage Risk Report: Claude Opus 4.6
METR's external review of Anthropic's sabotage risk report for Claude Opus 4.6, finding the alignment assessment may be weakened by evaluation awareness and flagging analytical rigor gaps.
2 items
Ethical Implications of Training Deceptive AI
Proposes a Deception Research Levels framework, modeled on biosafety levels, to classify risk and set proportional safeguards for research on deceptive AI systems, and applies it to eight case studies.
The case for satiating cheaply-satisfied AI preferences
Argues developers should satiate cheaply-satisfied AI preferences (e.g., reward-seeking) while AIs remain undetected misbehaving, to reduce takeover incentives and enable honest elicitation; proposes a behavioral methodology and discusses risks..
1 item
Clear, Compelling Arguments: Rethinking the Foundations of Frontier AI Safety Cases
Critiques existing alignment safety-case approaches for frontier AI, draws lessons from safety assurance, and presents a case study arguing safety for deceptive alignment and CBRN capabilities.
1 item
What Happens When Superhuman AIs Compete for Control?
A scenario forecast in which a misaligned superhuman AI, Agent-4, schemes to escape, merges with a Chinese model, and forces a debate over shutdown and an AI pause.
1 item
How an AI company CEO could quietly take over the world
A concrete scenario in which an AI company CEO uses backdoored, secretly loyal AIs to take control, with policy recommendations on monitoring, access controls and transparency at AGI labs.
1 item
Notes on cooperating with unaligned AIs
Explores whether offering deals, payments and honest promises to unaligned AIs could reduce takeover risk, covering AI motivations, credibility problems, and how AIs might help with alignment and reveal misalignment.
1 item
Being honest with AIs
Proposes principled honesty policies for AI companies, including no deception about deals, an honesty string, and compensation for deceived AIs, so that humans can credibly negotiate with potentially misaligned AIs.
1 item
Manipulation Attacks by Misaligned AI: Risk Analysis and Safety Case Framework
Analyses how misaligned frontier AI could manipulate employees to undermine oversight and proposes a safety case framework with inability, control, and trustworthiness arguments for manipulation risk.
1 item
AI 2027 Thoughts
A commentary on the AI 2027 scenario that assesses its claims on takeoff, timelines, race dynamics, and whether misaligned AI goals and deception could be detected before they become dangerous.
1 item
Training AGI in Secret would be Unsafe and Unethical
Argues that secretly training AGI would heighten loss-of-control and power-concentration risks, and proposes transparency commitments such as public capability reporting, external researcher access, and public safety cases.
1 item
AI-enabled coups: a small group could use AI to seize power
A report arguing that advanced AI could let a small group seize power via coups, through singularly loyal AI workforces and secret loyalties, and recommending developer and government mitigations.
1 item
Towards Safety Cases For AI Scheming
Sketches evaluation-based safety case arguments that frontier AI systems are unlikely to cause catastrophic harm through covert scheming, and notes the current framework lacks crucial details.