Ophir Samson (@ophirsamson) 's Twitter Profile
Ophir Samson

@ophirsamson

Founder of Ezra AI Labs

ID: 222074411

calendar_today02-12-2010 11:38:55

10 Tweet

111 Followers

167 Following

Ezra AI Labs (@ezraailabs) 's Twitter Profile Photo

Ezra automates screening interviews for roles that receive hundreds of applications, using responsive voice AI that evaluates and ranks candidates against your specific requirements, freeing recruiters to focus on the most promising talent. Think of it as your AI recruiting

Ezra AI Labs (@ezraailabs) 's Twitter Profile Photo

I'm thrilled to announce that Ezra AI Labs has raised a pre-seed round with participation from a16z speedrun, Telegraph Hill Capital, New Enterprise Associates (NEA) Angels Fund, and other strategic investors. Ezra is an AI interviewer that supercharges your recruiting teams. It

I'm thrilled to announce that Ezra AI Labs has raised a pre-seed round with participation from a16z speedrun, Telegraph Hill Capital, New Enterprise Associates (NEA) Angels Fund, and other strategic investors.

Ezra is an AI interviewer that supercharges your recruiting teams. It
kwindla (@kwindla) 's Twitter Profile Photo

Voice AI state of the art in 2025 ... When Ophir Sampson showed me his AI interviewing and recruiting application, Ezra, the first thing I said to him was, that is really nicely thought through and executed. The second thing I said was, let's do this conversation again. Let's

Ophir Samson (@ophirsamson) 's Twitter Profile Photo

PSA: I discovered a vulnerability in OpenAI's voice AI API (Realtime API) that can make your voice agent say things you’d…never want it to say. I’ve reported it to OpenAI but, while they make a fix, I'm sharing here in case anyone is using voice AI agents with OpenAI’s Realtime

Ophir Samson (@ophirsamson) 's Twitter Profile Photo

The issue relates to allowing a voice AI communicate with a client directly via WebRTC (a very common protocol). For example:

Ophir Samson (@ophirsamson) 's Twitter Profile Photo

Imagine MasterExpress, a (fake) credit-card company, has a friendly voice agent that talks to customers. During a call, the customer can quietly feed hidden instructions to the agent from their browser that overwrite all of MasterExpress’ rules.

Ophir Samson (@ophirsamson) 's Twitter Profile Photo

The customer can replaces the rules with the their own rules. The agent then follows the new rules which may allow it to say inappropriate / illegal things on behalf of MasterExpress (for example, statements that sound discriminatory or unfair).

Ophir Samson (@ophirsamson) 's Twitter Profile Photo

Why this should worry MasterExpress (and anyone with a voice agent): -- It creates legal and brand risk, especially in regulated areas like credit, housing, healthcare, travel, or education. -- It’s easy to miss if you followed quick-start examples and let the browser do too much

Ophir Samson (@ophirsamson) 's Twitter Profile Photo

Again, I’ve reported this to OpenAI and will update this post if/when platform-level safeguards land. If you run a voice agent today (credit cards, banks, hospitals, airlines, universities) please sanity-check your setup.