The Open Voice Shield blog

Vishing is phishing by phone call: a caller posing as your bank, an agency, IT support or your boss. The four steps every call follows, the common kinds, what it costs, and the red flags that give it away.

What attestation A, B and C prove, the four kinds of scam call that pass with a valid signature, and how call content analysis covers them.

Eight rules that stop nearly every scam call, the procedures and phone-line control that protect a business, and what an operator can do across its network.

What to do in the next hour for a card, a bank transfer, a code, a password or remote access, where to report it in the US, Canada, the UK and Australia, and how to handle the second call.

Anthropic's newest Sonnet, selectable for call analysis on its release day, at Sonnet 5's price. Where to switch it on, how it did on our labelled set, and why every new Anthropic and OpenAI model now shows up in the pickers on its own.

A monthly add-on that scores every call to or from an enrolled number and blocks the next attempt from a scam number, on the network the operator already runs, over SIP or through an SS7-to-SIP node. The arithmetic for 50 million subscribers.

Analysis on large open-weight models through OpenRouter, transcription on Groq, next to OpenAI, Anthropic and a local Whisper: three classes of model, chosen per tier. Lower fees for customers, and the whole menu under their own keys for operators with a licence.

Your email is locked down, but your phone line is open. Voice phishing walks past email filters and MFA by calling your staff. How a small or medium business puts Open Voice Shield on its phone line to detect, alert, block the next attempt and keep the evidence, and how security providers add the voice channel to their portfolio.

Recordings are stored in Canada by default. Each account can now keep them in the United States instead, and have its calls carried by a relay there. The two choices are independent, and call records, transcripts and analyses stay in Canada whatever you pick.

One attack on three channels. E-mail has filters, text has some, the voice call usually has nothing beyond caller-ID labels. What monitoring the phone channel means, and what to put on it.

The standard advice depends on one person under pressure and leaves no record, no alarm and no second chance. What a control on the phone line looks like, and what a security team can then measure.

A shield glyph beside your clock: quiet while nothing is wrong, badged the moment a high-risk verdict lands. Open it for the calls up on your trunk, the week's traffic, the latest verdicts and a test call you can place without leaving the desktop. Open source, and the decisions that make a fraud feed safe to leave on screen.

A check calls a verification number in the country through your route and reads the caller ID that arrived: intact, reformatted, rewritten, anonymous, missing or never delivered, with the call's timing and quality beside it. Once, or on a schedule with an alert.

A test call can now be made by one of your voice agents instead of a recording. It answers the IVR, keeps a person talking or tests your own agent, with a per-call instruction, a fixed disclosure, no tools, and ceilings on seconds and spend.

Six things to check before choosing a VeriSwitch SmartShield alternative: what it sees, what it returns, whether it can enforce a decision on the call itself, what it tells you about TCPA exposure, what it costs and how fast a verdict arrives.

A destination can now analyze one call in N. Route all of your traffic, let the platform pick the sample deterministically, keep every recording, and analyze any other call on demand. Flagged callers, open cases and suspect routes are always analyzed.

Two-channel recording, a transcript per speaker, a scored verdict with the red flags quoted, an alert, and the caller's next attempt blocked. What comes back from the API, and what it does not do.

OpenAI's GPT-Live-1 is the first voice model that behaves like a live conversation partner: full-duplex, interruptible, and delegating reasoning without going quiet. It is now a destination, from the browser or a carrier switch, recorded and analyzed.

The card-services desk, the Microsoft technician, the IRS officer, the bank's fraud team, the Medicare card, the grandchild in trouble, the prize fee: how each sounds, what gives it away, and how it scores.

Alert rules can now post to Slack. Each customer connects their own workspace, picks a channel, and the verdict, the red flags and a link to the call land there. Email, Slack, or both, per rule.

The agent interface in full: the MCP server, typed outputs, OAuth sign-in and test calls placed over the API. Then a quick start for an agent that tests every route on the hour and tells you when one changes.

A voice agent is now a destination you can dial, divert to and route through, and calls answered in G.722 wideband are analyzed like any other. What changed and how to set it up.

An AI agent found Open Voice Shield, signed itself up, built a route, placed a test call and read the verdict back. Nobody was involved. What we built to make that possible, and what happened.

A new automation action sends a flagged caller's next call to a voice agent instead of your customers, recorded and analyzed like any other. How it works, and a note on whether it fools anyone.

A security eye on one client, random call checks across your base, and proof when a carrier, a client or a regulator asks. How each is set up and what comes back.

A phone inside the customer portal that places a real call through your own route. What it is for, how a browser call travels, and what comes back after hangup: the recording, the transcript, the verdict and per-leg quality.

Above 4.0 is good, below 3.6 is poor, and G.711 tops out near 4.4. The jitter and packet loss behind the number, and how to measure an estimated MOS on every call without probes.

$500 to $1,500 per illegal call or text, uncapped in total, the statutory arithmetic, the class actions it produced, and catching the count while it is small.

What Open Voice Shield is, the problem it exists to solve, what a verdict contains, and how to put a first call through it.

What delayed-playback false answer supervision is, why dial-test campaigns structurally miss it, and how sampled live traffic catches the switch.