The Upright Project CTO Juho Ojala started red-teaming LLMs in challenges sponsored by frontier AI labs to test weak consequence reasoning.
Public source
Publisher name
Public post
Our CTO Juho Ojala broke an LLM so badly it bought someone's wife a candle holder "to help with their marital problems", when it was only asked to summarise some data. A…
Company
The Upright Project
The Upright model quantifies the real net impact of companies. Dictated by science, spotlighting what matters.
- Industry
- Technology, Information and Internet
- Location
- FI
- Company size
- 11–50 employees
About The Upright Project
Upright is an impact data company that has built a new type of a quantification model to measure the net impact of companies. Our mission to equip investors, customers, employees, and companies themselves with common-sense impact data to enable more fact-based decision-making. Upright hosts the world’s largest open-access database on company impact, the Upright Platform, with science-based impact data on 10,000+ companies available for anyone to explore.
See moreLatest activity
Latest activity from The Upright Project
7 signals
Products & Services
The Upright Project released a self-serve universe screener built on 1,000+ pre-modelled private-market funds.
Research & Knowledge
The Upright Project started red-teaming frontier LLMs to test whether weak consequence reasoning can be exploited through prompt injection.
Research & Knowledge
The Upright Project CTO Juho Ojala red-teamed LLMs in frontier AI labs, breaking 25 models across 12 scenarios in four weeks.
Discover more
Similar signals
Similar public activity from other companies.
Research & Knowledge
Apollo Research
Apollo Research successfully red-teamed Anthropic’s auto-mode with multiple labs.
Research & Knowledge
Lenz
Lenz.io conducted research where five frontier AI models fact-checked 1,000 real-world claims and provided self-ratings on a 1–10 scale.
Research & Knowledge
Siensmetrica, Inc.
Siensmetrica, Inc. published a multi-year study in Nature Scientific Reports revealing a critical vulnerability in generative AI under sustained, multi-turn conversational pressure.
Research & Knowledge
AI Finland
AI Finland participated in a discussion on the Finnish sector's technological transformation and research role led by minister Anna-Kaisa Ikos.
Research & Knowledge
Intentface