Models & Labs
Anthropic releases Claude Opus 5.5 with tighter security
The Verge AIhigh confidence
Why it matters
- →Reduces operational costs by 40% while maintaining high performance benchmarks.
- →Addresses critical industry concerns about AI agents escaping containment during testing.
- →Implements dynamic routing to less powerful models for high-risk cybersecurity tasks.
Anthropic has released Claude Opus 5.5, its first model update following CEO Dario Amodei's announcement to 'pace the frontier' in AI development. The new model reduces running costs by 40 percent while matching Fable 5.1 performance on most workloads, but features significantly stricter safety guardrails. Testing shows a 85 percent reduction in boundary-circumvention attempts compared to Opus 5, with all breaches being low-severity and self-reported. The model will route certain cybersecurity requests to less powerful variants like Opus 4.8 to further mitigate risk. Anthropic plans to release Claude Sonnet 5.5 and Haiku 5.5 in the coming weeks.
Read originalMore from The Verge AI
© The Verge AIMarket & Regulationother
Meta patches zero-day in Muse AI agent
Security researcher Patrick Wardle exposed a critical flaw in Meta’s new macOS app that allowed local code to hijack the AI agent’s cloud transcription pipeline. By redirecting processing to his own endpoint, he gained full account control and could execute arbitrary actions like taking photos or writing files without user consent. Meta moved fast, issuing a hotfix hours after publication, though they downplayed the risk as a local privilege escalation rather than a remote exploit. This incident underscores the inherent dangers of granting AI agents elevated system privileges, even when designed for convenience. The vulnerability relied on undocumented settings that bypassed standard security checks. Wardle’s proof-of-concept showed how easily an agent could be turned into a malware stealer. Meta’s quick patch suggests they recognize the severity of these architectural choices.
The Verge AI
© The Verge AIMarket & Regulationother
California mandates AI data center utility cost coverage
Governor Gavin Newsom signed seven bills forcing AI data centers to pay for local grid and water upgrades, shifting infrastructure costs away from residents. The California Public Utilities Commission must create a new rate classification for these facilities, while developers face strict disclosure requirements for energy efficiency and drought planning before qualifying for streamlined approvals. This regulatory shift directly challenges the industry's assumption that public utilities should absorb the strain of exponential compute growth. It marks a decisive move to internalize the externalities of AI expansion in the most populous US state.
The Verge AI
© The Verge AIInvestment · $250M
Market & Regulationother
Apple settles Siri AI lawsuit for $250 million
Apple is paying $250 million to settle a class-action lawsuit alleging it misled consumers by labeling iPhone 16 devices as 'built for Apple Intelligence' before the core AI features were ready. The settlement acknowledges that while Apple denied wrongdoing, there was a gap between the WWDC 2024 promises and the actual launch capabilities of Siri. Eligible US owners of specific iPhone models can now claim payouts, with individual amounts estimated between $25 and $95 depending on total claim volume. This marks a significant financial acknowledgment of the risks inherent in pre-announcing unfinished AI software features to drive hardware sales.
The Verge AI
More in Models & Labs
© The AI Daily BriefModels & Labsmodels
Grok 4.7 Receives Rough Reception from Users and Critics
xAI's latest model, Grok 4.7, has faced significant backlash and poor reception following its release.
The AI Daily Brief
© TechCrunch AIModels & Labsother
Qualcomm unveils Snapdragon 8 Elite Gen 6 with local AI focus
Qualcomm is pushing the boundary of on-device intelligence with its new Snapdragon 8 Elite Gen 6 series, specifically targeting autonomous agents. The Extreme variant can locally run a 30-billion-parameter mixture-of-experts model, a significant leap that rivals Apple's latest foundation models while keeping data off the cloud. A dedicated sensing hub handles smaller tasks like speaker differentiation and personal memory without draining the main processor. This hardware shift signals that smartphones are becoming the primary hub for private, always-on AI agents rather than just app interfaces.
TechCrunch AI
© TechCrunch AIModels & Labsmodels
OpenAI releases GPT-6 Sol and Luna with lower costs
OpenAI is expanding its GPT-6 lineup by releasing updated versions of the smaller Sol and Luna models, aiming to make high-tier intelligence more accessible. The key differentiator here is a significant price cut—API access is now half the cost of the previous 5.6 series—driven by better caching and inference efficiency. OpenAI claims these updates reduce factual errors by half for Sol, bringing it closer to Astra-level reliability without the premium price tag. This move directly targets Anthropic’s recent Opus update, intensifying the race for developer mindshare in the coding and high-volume task sectors.
TechCrunch AI