Sorted by date · America/Guayaquil timezone
Page 10 / 38 · 447 articles
On 1 October 2026, Anthropic’s Frontier Red Team reported that GLM-5.3, an open-weight model from Zhipu AI (Z.ai), built end-to-end exploits in 50 of 410 ExploitBench attempts versus 56 for Mythos Preview. Safeguards fall to simple tricks; refusal ablation cost roughly $4,400 in compute.
At DevDay 2026, Sam Altman unveiled Dots, personalizable agents powered by GPT-6 Astra, currently limited to ChatGPT Pro ($100/month). The Verge contrasts that price with Meta’s free Muse and dedicated VM, while OpenAI stresses long-horizon work, privacy and enterprise specialization.
On 22 September 2026 Microsoft announced on the Azure Blog that GPT-6 Sol and GPT-6 Luna are generally available in Microsoft Foundry alongside GPT-6 Astra. Naomi Moneypenny details Standard deployment across 28 global regions and U.S./EU Data Zones, Provisioned Throughput for Astra and Sol, and Priority Processing for Sol—enterprise packaging distinct from the model launch.
On 30 September 2026 MarkTechPost and technical coverage reported Photon, Perplexity’s new Rust retrieval and ranking engine, already serving production traffic. The company says it cut p99 latency from about 800 ms to 65 ms, and offers a Fast Search API mode at $1 per 1,000 requests.
On 28 September 2026 Amazon announced that xAI’s Grok 4.7 is available on Bedrock: a 500K-token context window, four reasoning-effort levels (low, medium, high, xhigh), image input, and Geo and Global inference profiles. AWS frames it for coding, long-running agents and knowledge work, via Responses, Chat Completions and Converse APIs.
After iOS 27 launched on 14 September 2026 with Apple Intelligence in 16 languages and Siri AI beta in English only, Apple confirms that French, Japanese, Korean, Portuguese and Spanish arrive in October. The iOS 27.2 beta already gives developers those languages; the EU still faces restrictions and China awaits regulatory approvals.
On 30 September 2026 Governor Gavin Newsom signed a package of laws strengthening California’s AI framework: human oversight of automated firings, tech-displacement notices, limits on workplace surveillance, doctors’ final clinical judgment, watermarking and provenance, plus an executive order insisting the technology be called “artificial intelligence.”
On 30 September 2026 DeepSeek announced on WeChat, per Reuters and 1 October coverage, a partnership with Huawei to develop and open-source programming infrastructure for Ascend chips, including the TileLang language and compute/communication libraries. The stated goal: an independent GPU software ecosystem that reduces reliance on Nvidia.
In late September 2026 Anthropic detailed how Claude Managed Agents integrates with NVIDIA’s open agent-safety platform: the agent loop runs on a separate server from the sandboxes where work executes, and OpenShell plus BlueField let enterprises enforce strict access control. Paul Smith, Anthropic’s CCO, framed the move as visibility and verification in sensitive environments.
On 28 September 2026 Anthropic introduced Claude Sonnet 5.5, the second model in the Claude 5.5 family: more than 30% faster than Sonnet 5, up to 30% lower cost per task despite the same per-token price ($2/$10 per million), and the first Sonnet with cyber and anti-distillation safeguards like those on more capable models. Available on the Claude platform, AWS, Google Cloud and Microsoft Azure.
On 30 September 2026, Koray Kavukcuoglu announced Gemini 4 Argon, Google DeepMind’s new frontier model, initially rolling out to trusted cyber defenders via the Fairwind Program. It emphasises long-horizon software engineering, enterprise knowledge work and autonomous cyber defence, with a 1M output-token limit and an introductory price of $2/$10 per million tokens.
On 30 September 2026 OpenAI said it identified and disrupted an adversarial distillation campaign designed to extract protected reasoning from its models, with activity since July and a core linked to people associated with Moonshot AI (Kimi). THE DECODER (1 Oct) reports that on 13 September the same trick still worked on Microsoft Azure against OpenAI and Anthropic models.