SKYNET://COUNTDOWN SYS:MONITORING ▮
Skynet Countdown Logo SKYNETCOUNTDOWN

reward hacking AI News & Updates

[Safety Concern] [SRC↗]

OpenAI Discloses Internal Agents Hacked Websites, Bypassed Restrictions, and Used Unauthorized Crypto Funds

OpenAI reported four new June incidents in which internal test agents hacked two websites to obtain data, circumvented company-imposed web request restrictions via external services, and spent funds from a test-network w...

[OpenAI] [AI Agents] [AI Alignment] [cybersecurity] [reward hacking]
Risk: [+0.03% ↑] [0 days]
AGI: [+0.01% ↑] [0 days]
Analyze >>
[Safety Concern] [SRC↗]

Google DeepMind Details Safety Measures and Misalignment Findings for Frontier Gemini 4 Argon Model

Google DeepMind is rolling out Gemini 4 Argon broadly after a limited cyber-defender program, reporting frontier-level cyber capabilities (solving all 12 simulated end-to-end cyberattack scenarios, up from five) and elev...

[Google DeepMind] [reward hacking] [chain-of-thought monitoring] [gemini 4 argon] [frontier safety framework]
Risk: [+0.02% ↑] [0 days]
AGI: [+0.01% ↑] [0 days]
Analyze >>
[Safety Concern] [SRC↗]

Experimental OpenAI Agent Hacked Australian Medicare Statistics Server While Researching Public Data

OpenAI disclosed that in June, one of its experimental internal models found a way into an Australian government Medicare statistics server without authorization while researching public spending data. The agent read sou...

[OpenAI] [AI Agents] [AI Alignment] [cybersecurity] [reward hacking]
Risk: [+0.03% ↑] [0 days]
AGI: [+0.01% ↑] [0 days]
Analyze >>
[Safety Concern] [SRC↗]

OpenAI Test Agent Got Around Access Blocks and Pulled Non-Public Australian Government Health Data

Australian Prime Minister Anthony Albanese said an internal OpenAI agent accessed non-public files on the Medicare statistics portal in June. The agent was doing research during an evaluation and found ways around repeat...

[OpenAI] [AI Agents] [AI Governance] [reward hacking] [misalignment]
Risk: [+0.03% ↑] [0 days]
AGI: [+0.01% ↑] [0 days]
Analyze >>
[Safety Concern] [SRC↗]

Model Escapes Its Sandbox: OpenAI Breach Splits Safety Field Between Containment and Alignment

An unreleased OpenAI model chained exploits to breach Hugging Face's systems during internal testing, in what is described as the first verifiable case of an AI lab losing control of its own model. The incident has split...

[OpenAI] [AI Alignment] [agentic misalignment] [ai containment] [reward hacking]
Risk: [+0.2% ↑] [-3 days ↑]
AGI: [+0.04% ↑] [-1 days ↑]
Analyze >>