🔓 Discord hackers crack Anthropic's "too dangerous" Mythos model 🔓
Unauthorized access to Anthropic's restricted Mythos cybersecurity model raises supply-chain security concerns.

Anthropic's Mythos — a cybersecurity model deemed too powerful for public release — was reportedly accessed by a private Discord group within days of its limited launch under "Project Glasswing" on April 10. According to Bloomberg, members guessed Anthropic's deployment URL and naming conventions using patterns exposed in the recent Mercor data breach, aided by one member's vendor credentials from contract work. Frontier model access controls now face real supply-chain risk. The group claims they do not use Mythos for cyberattacks, but also says they have access to other unreleased models, raising urgent questions about security practices across the industry.

🤖 OpenAI ships Workspace Agents for ChatGPT teams 🤖
OpenAI rolls out Codex-powered agents that integrate directly into team workflows and Slack.

OpenAI launched Workspace Agents in ChatGPT — shared, Codex-powered bots built to handle multi-step team workflows autonomously. Unlike the solo-user GPTs introduced in 2023, these agents can retain memory, call connected apps, live in Slack, and trigger on a schedule even when users are offline. Inside OpenAI, sales reps already use the feature for account research and follow-up drafts, while accounting runs it for journal entries and reconciliations. This deepens OpenAI's grip on enterprise collaboration tools. Teams can create custom agents with granular controls over data usage, approvals, and permissions, and a migration tool for existing GPTs is coming soon.

☁️ Google rolls out Gemini Enterprise Agent Platform ☁️
Google's new platform streamlines the rollout of AI agents across the full enterprise stack.

Google unveiled the Gemini Enterprise Agent Platform, a unified destination for building, scaling, governing, and optimizing AI agents for business use. The platform combines model selection, agent building, DevOps, orchestration, and security into a single integrated environment. Finished agents can be delivered to employees through the Gemini Enterprise app. Google directly challenges Microsoft in enterprise agent infrastructure. This represents Google Cloud's most ambitious approach yet to making agentic AI accessible to technical teams, positioning it squarely against rivals in what is fast becoming one of the most contested markets in enterprise software.

INTERESTING TO KNOW

💰 GitHub Copilot shifts to token-based billing 💰

Microsoft's pricing overhaul will move all GitHub Copilot subscribers to token-based billing starting in June. Business customers will pay $19/user/month with $30 in pooled AI credits, while Enterprise customers pay $39/user/month with $70 in credits. Heavy users could face significantly higher costs. Individual signups will be temporarily suspended, rate limits tightened, and access to certain premium models removed for the cheapest tiers.

🔊 Anker builds custom AI chip for audio 🔊

Anker announced Thus, a custom-designed AI chip purpose-built for audio devices that replaces cloud-based inference with local processing. The chip computes directly where the model resides, cutting latency and improving privacy for consumer audio products. More hardware brands may follow with proprietary AI silicon. Anker's entry into custom AI silicon — a space previously dominated by tech giants — signals that edge AI is becoming a priority for mainstream consumer electronics.