Release Notes - Platform
These are the release notes for updates to the Language IO Platform.
The Language IO Platform encompasses the core technology and services, feature ecosystem, and related configurations that customers may interact with through our CRM channel integrations and API. The Platform also includes our Customer Self-Service Portal.
April 28, 2026: 7.22
New
Introducing Large Language Models (LLMs)
Following extensive testing to curate the highest-quality translation models for our customers' specific translation needs, Language IO has integrated Large Language Models (LLMs) into our translation pipeline.
With Language IO's multi-model solution, we carefully evaluate each translation model's performance, and our solution dynamically selects the best model for each of your translations.
Language IO's LLMs deliver:
- Superior translation quality and fluency with enhanced natural phrasing, contextual intelligence, and localized formatting.
- Advanced glossary and morphology handling that improves glossary utility and reduces maintenance.
- Massive scale and formality support that expands language support and enhances stylistic control.
- Zero-Trace policy consistency that ensures full compliance with existing privacy policies. Customer data is never used to train models.
For a detailed breakdown of LLM strengths and capabilities, see Large Language Models for Translations.
Improved
API Key Rotation With Rolling 14-Day Expiration Window
Customer Self-Service Portal admins can now rotate their API key safely by generating a new key without immediately invalidating their existing key. Once a new key is generated, the old key remains valid for 14 days, giving you time to swap in the new key where needed.
For more information, see Managing your API key.
March 24, 2026: 7.20
Fixed
Translated Articles Failed to Return to the Customer Self-Service Portal
A bug was found where, when users submitted an article project, whether in staging or production, the project would complete in XTM but would fail to return to the CSSP.
This is now fixed, and completed projects are returned as expected.
March 10, 2026: 7.19
New
Introducing the Toxicity Shield
Language IO is proud to announce a new feature that will help agents deal with potentially problematic end users: the Toxicity Shield.
The Toxicity Shield feature is an automated, LLM-based moderation tool that assesses the toxicity of a message to assign it a score, then take certain actions or not, as applicable, based on this score.
The Toxicity Shield runs when the message is ready to be delivered to its recipient (this means after the translation, if a translation was performed). How it works can be broken down into two parts:
- Assessing and assigning a toxicity score to a message.
- Use this score as a trigger for certain actions:
- displaying an icon along with a chat message to indicate the "temperature" of the message
- Flagging, rewriting, or even blocking a message depending on a sliding threshold.
The Toxicity Shield runs on both inbound (end user to agent) and outbound (agent to end user) messages.
- When a message from the end user is rewritten, the agent can click on the "eye" icon to view the original message. The translation of this original message is displayed between the customer message and the rewritten message.
- When a message from the agent is blocked, this is clearly indicated in the user interface, so that they know that they need to take a follow-up action:
- If the message from the agent is only blocked, the agent gets a prompt to rewrite the message.
- If the message from the agent is blocked and rewritten, the rewritten message shows as a suggested reply, which the agent can either accept and send, edit and send, or reject and send the original.
The Toxicity Shield is currently available in Salesforce Enhanced Chat from v1.21. In the meantime, CSSP admins can configure and preview the Toxicity Shield in the CSSP.
February 10, 2026: 7.18
New
Timeouts are now Configurable by Consumer and Channel
Translation timeouts can now be configured at the consumer and Channel (content type) levels. With this improvement, customers can now have fine-grained control over performance behavior so they can balance latency and translation completeness depending on the channel (for example, setting a 3-second timeout for Chat, but a separate 10-second timeout for Case).
For more information about how to request these configurations, submit a Support request.
Fixed
Custom Locales not working for Attachment Translation
A bug was found where Salesforce Case customers who had configured custom locales would not be able to perform translations of attachments.
This is now fixed, and Case users should now be able to translate attachments to custom locales as expected.
Bolded Text introduced Extra Spaces
A bug was found where, when a text contained portions in bold, Machine Translation vendors would add extra spaces before and after the bolded section. As a result, this could cause further formatting issues.
This is now fixed.