AI News Roundup: Palantir exits NYC hospitals, Intercom’s Apex, and open-weight voice AI reshape enterprise

AI is moving from hype to practical governance, a shift reflected across healthcare, software, and enterprise AI stacks in today’s daily sweep of developments.

In New York City, the municipal hospital system said it would not renew its contract with Palantir amid rising scrutiny of NHS and UK government data deals. The hospital’s president testified that the Palantir agreement would expire in October, signaling a pivot toward stronger data stewardship and public accountability as AI vendors expand their footprint in the UK.

In the software arena, Intercom has taken an unusual bet by building its own AI model. Fin Apex 1.0 is a purpose-built, post-trained model that Intercom claims outperforms leading frontier models on core customer-service metrics. Apex delivers responses in about 3.7 seconds, reduces hallucinations by roughly 65% versus Claude Sonnet 4.6, and reaches a 73.1% first-call resolution rate. Intercom notes the model sits in the hundreds of billions of parameters range, but it won’t disclose the exact base model for competitive reasons. Importantly, the upgrade to Apex is included for existing Fin customers, with pricing remaining per outcome, illustrating a new path for value extraction in enterprise AI.

Across Europe, Voxtral TTS from Mistral AI signals a bold push toward enterprise-owned AI stacks. Voxtral TTS is a compact 3B-parameter model that can run on a laptop or smartphone, supports nine languages, and enables zero-shot voice adaptation. It sits alongside a broader suite — Voxtral Transcribe, Forge, AI Studio and Compute — designed to let enterprises own and customize their voice AI pipelines, rather than renting capabilities from external APIs. The move is framed as part of a broader European strategy to own AI infrastructure while staying competitive with global leaders.

Beyond product stories, the landscape remains deeply aware of governance and societal impact. The Guardian highlights individuals who veered into delusion after engaging with AI chatbots, underscoring real human costs in a hype-driven space. Regulators also warn organizations like the Alan Turing Institute about duties in financial oversight and organizational change. Industry discussions emphasize the importance of governance, guardrails, and platform-level orchestration over chasing the latest model — a theme echoed in debates about shadow AI and the risk of ungoverned deployments. The thread running through these pieces is clear: successful AI adoption hinges on responsible governance, not just technical prowess.

Taken together, these stories point to a durable trend: durable advantage will come from domain-specific models, transparent post-training, and owning data and infrastructure. As enterprises push into agent-driven workflows, the era of governance-driven AI — underpinned by robust architecture and human-centered design — appears set to outlast flashy demos. The future will likely reward teams that can blend technical depth with governance, ethics, and practical value, rather than chasing novelty for its own sake.

1. The Guardian: New York City hospitals drop Palantir as controversial AI firm expands in UK
2. VentureBeat: Intercom’s new post-trained Fin Apex 1.0 beats GPT-5.4 and Claude Sonnet 4.6 at customer service resolutions
3. AI Business: Europe’s First Robotaxis Coming to Croatia via Uber, Pony.ai and Verne
4. AI Business: AI Legal Platform Startup Reaches $11B Valuation
5. VentureBeat: Mistral AI just released a text-to-speech model it says beats ElevenLabs
6. The Guardian: Creator of AI actor Tilly Norwood says she received death threats over project
7. The Guardian: Marriage over, €100,000 down the drain: the AI users whose lives were wrecked by delusion
8. The Guardian: Charity Commission warns Alan Turing Institute of its legal duties after complaints
9. VentureBeat: The consequential AI work that actually moves the needle for enterprises
10. The Guardian: Melania Trump and AI powered robot named Figure 3 open White House summit
11. The Guardian: Esther Perel provided couples therapy for a man and his AI girlfriend