Apple built an internal chatbot app called Veritas to test a major Siri upgrade

Source Cryptopolitan

Apple has developed a ChatGPT-style iPhone app to test major upgrades coming to Siri next year, according to Bloomberg. The app, called Veritas, is designed for internal use only.

Employees are using it to test whether the new version of Siri can finally compete with smarter assistants like Google’s Gemini and OpenAI’s ChatGPT. The current plan is to launch the updated Siri as early as March 2025, after multiple delays and failed rollouts.

Veritas was built by Apple’s internal AI division to help engineers test Siri’s new abilities more efficiently. With this app, employees can simulate how Siri handles private data, like emails, music, and photos, and test if it can carry out actions like editing or searching.

Apple has no plans to release Veritas to users. It exists to let developers push the new Siri software without waiting for full OS releases.

Apple tests new large language model with Veritas

Veritas is running on a new backend system called Linwood, which uses a large language model trained by Apple’s Foundation Models team along with tech from an unnamed third party.

It works like popular chatbots; users can have multiple conversations, go back to earlier chats, ask follow-up questions, and interact with the system in longer exchanges. It’s all designed to test how well Siri can handle real human input.

The updated Siri was originally supposed to launch in spring 2024, but Apple had to delay the rollout after internal tests showed the features failed about one-third of the time. That failure triggered a leadership shakeup.

Apple’s AI chief, John Giannandrea, and several key deputies were sidelined, and Robby Walker, the executive who previously led the Siri team, is leaving the company in October. After losing control of Siri, Robby started a new internal team called AKI, short for Answers, Knowledge, and Information.

That group is now building new AI-based search features for Siri. Apple has already started building more features on top of Siri. The new version will be able to recognize what’s showing on a user’s screen and act on it; something the current Siri can’t do.

It’ll also allow users to move around their devices using just voice commands. Veritas helps test all that. It also includes tools that let engineers gather information from the internet and see AI-generated summaries.

Apple explores deals with AI firms while redesigning Siri

Apple isn’t doing this alone. It’s been shopping around for external help. Earlier this year, it held talks with OpenAI about using its models in Siri. Then came discussions with Anthropic, the makers of Claude. Now, Apple is deep in talks with Google about integrating a custom version of Gemini into Siri’s infrastructure. No deal has been finalized, but Apple is keeping its options open.

While all this testing is going on, Apple is also preparing a visual redesign for Siri. That new interface is expected before the end of 2025. The company is working on more AI features for HomePod, Apple TV, and even smart home devices that haven’t been announced yet. All of this is part of Apple’s broader strategy to stay in the AI race.

Despite all the internal activity, Apple isn’t planning to release Veritas to customers. That’s consistent with what Apple executives have said publicly. In June, Apple software chief Craig Federighi told Tom’s Guide that chatbots like Veritas are powerful, but “remain not our primary goal.” When asked about releasing something like ChatGPT, Craig said, “Time will tell where we go there.”

Siri’s delayed launch and the failed spring rollout forced Apple to rethink its entire AI strategy. The company knows it’s behind. When it launched the iPhone 17 this month, it didn’t even mention its in-house AI system. But Apple isn’t dropping the fight.

The Veritas app is being used to pressure-test every feature ahead of the 2025 release. Whether or not Siri catches up to its competitors, Apple is betting it still has time to fix its voice assistant before AI becomes the deciding factor in smartphone sales.

KEY Difference Wire helps crypto brands break through and dominate headlines fast

Disclaimer: For information purposes only. Past performance is not indicative of future results.
placeholder
Anthropic files for a $2 trillion IPO — the $80.7 billion loss is the decoy, the $450 billion compute bill is the riskAnthropic publicly filed its S-1 on October 26, revealing $16.29 billion of first-half revenue (+1,304% year over year) and a $80.7 billion net loss — of which $72.4 billion is a non-cash convertible-note charge. The real number is the $450 billion of signed, mostly non-cancelable compute commitments. Here is what the filing says, why NVDA, ARM and MU are the tradeable read-through, and the conditions that decide whether the AI trade passes its first public valuation test.
Author  Irene Q.
Oct 27, Tue
Anthropic publicly filed its S-1 on October 26, revealing $16.29 billion of first-half revenue (+1,304% year over year) and a $80.7 billion net loss — of which $72.4 billion is a non-cash convertible-note charge. The real number is the $450 billion of signed, mostly non-cancelable compute commitments. Here is what the filing says, why NVDA, ARM and MU are the tradeable read-through, and the conditions that decide whether the AI trade passes its first public valuation test.
placeholder
Gold Price Analysis Today | Forex Gold Analysis | XAU/USD Analysis - October 29, 2026Get the latest gold price analysis, forex gold insights, and XAU/USD analysis for October 29, 2026. Read on for the details.
Author  Mitrade
Oct 29, Thu
Get the latest gold price analysis, forex gold insights, and XAU/USD analysis for October 29, 2026. Read on for the details.
placeholder
The Fed held at 3.75%–4.00% — and two officials voted to hike. December is now 85% pricedThe FOMC held by a 10–2 vote, with Governor Cook and Cleveland Fed President Hammack dissenting for a 25bp hike. Chair Warsh offered no forward guidance and no dot plot, the 2-year yield fell 8bp to 4.70%, and the dollar slipped to 101.84 even as markets moved December hike odds to 85%–87%. Tonight's Q3 GDP and September core PCE are the next referee.
Author  Irene Q.
Oct 29, Thu
The FOMC held by a 10–2 vote, with Governor Cook and Cleveland Fed President Hammack dissenting for a 25bp hike. Chair Warsh offered no forward guidance and no dot plot, the 2-year yield fell 8bp to 4.70%, and the dollar slipped to 101.84 even as markets moved December hike odds to 85%–87%. Tonight's Q3 GDP and September core PCE are the next referee.
placeholder
Gold Price Analysis Today | Forex Gold Analysis | XAU/USD Analysis - October 30, 2026Get the latest gold price analysis, forex gold insights, and XAU/USD analysis for October 30, 2026. Read on for the details.
Author  Mitrade
Yesterday 02: 21
Get the latest gold price analysis, forex gold insights, and XAU/USD analysis for October 30, 2026. Read on for the details.
placeholder
US GDP grew 3.4% and core PCE held at 3.1% — December is now a hike trade, not a cut tradeUS Q3 GDP rose 3.4%, above the 3.1% estimate, while September headline PCE hit 3.6% and core held at 3.1%. December hike odds are near 90%; inventories, AI capex and energy explain why.
Author  Irene Q.
20 hours ago
US Q3 GDP rose 3.4%, above the 3.1% estimate, while September headline PCE hit 3.6% and core held at 3.1%. December hike odds are near 90%; inventories, AI capex and energy explain why.
goTop
quote