Reading

Anthropic released Opus 5.5 on September 22, cutting output token pricing to $20 per million tokens from $25 for the previous model, a 20% drop.

The company calls it the strongest-performing model it has tested and says it outpaces its own larger Fable model on many benchmarks. Those results are all vendor-run; METR and other outside groups did pre-release safety evaluation only, and no independent performance verification exists yet.

Anthropic says the model is comparable to Mythos in biology and cybersecurity capabilities, so it carries the same usage safeguards as Fable. Sonnet 5.5 and Haiku 5.5 are promised in the coming weeks; whether the price cut extends to mid-tier models is what buyers should watch.

The hacking group ShinyHunters claims it breached FBI-related systems and stole data on "all FBI employees and applicants." The FBI has not responded to a request for comment.

The group gave 404 Media a sample covering about 5,000 FBI employees, including addresses, phone numbers, dates of birth and, in some cases, spouse details. 404 Media ran some sample phone numbers through OSINT Industries and the names did correspond. The hackers say they used a zero-day in Oracle PeopleSoft, then reached AWS GovCloud servers and exfiltrated roughly two to three terabytes; they also defaced the FBI jobs site on Tuesday, which currently shows the applicant portal as unavailable.

The boundary: the core facts so far are the hackers' own claims, the sample was only checked for name-phone correspondence, and the "all employees" scale is unconfirmed by the FBI. The group says the motive is not financial; watch what it does next.

AMD rose nearly 10% to close at $615.52 on September 21, pushing its market capitalization above $1 trillion for the first time.

The stock is up about 30% over the past 30 days. Intel shares rose more than 12% the same day, and Reuters noted the move was a broad rally in AI-adjacent hardware stocks rather than an AMD-specific event.

Fast Company attributes the surge to investor optimism about AI infrastructure spending, citing the popularity of Meta's new AI app Muse. That causal story is the article's interpretation; AMD disclosed no new orders or results that day.

Broadcom CEO Hock Tan says Anthropic's renewed call to slow frontier model development has not shaken his outlook to ship roughly $350 billion of AI semiconductors to customers over two years.

Per MarketBeat, asked whether the AI slowdown debate gave him any pause, he answered "No, not in the least." The remark follows a Sept. 14 selloff in which Broadcom fell 4.8% after Dario Amodei's pacing post.

His reasoning: both training and inference demand stay strong and durable, and Broadcom's custom chips for Alphabet, OpenAI and Anthropic lean toward inference. This remains management's own statement; whether shipments materialize will show up in orders and earnings, not in the forecast.

Xiaomi on September 22 released and open-sourced its MiMo-V2.6 series, comprising Pro, Flash and a faster Pro-UltraSpeed variant, with text, image, video and audio input and a 1-million-token context window.

The company says Pro scored 46.32 on the Artificial Analysis Intelligence Index, the highest among open-weight models at launch, while still trailing closed flagships such as Claude Fable 5.1 and GPT-6 Astra; agentic scores come close to or slightly exceed Claude Opus 5. All of these figures are Xiaomi's own tests, with no independent reproduction yet.

Pricing carries over from MiMo-V2.5: Pro costs $0.435 per million input tokens and $0.87 per million output, Flash $0.14 and $0.28, while Pro-UltraSpeed trades 20-times-faster generation for $4.35 and $8.70. The models are listed on OpenRouter and Xiaomi's AI Studio.

Databricks announced on September 22 that the Genie One MCP server is now generally available to all users, ending its preview phase.

The interface lets any MCP-compatible agent — Claude, ChatGPT, Cursor or internal tools — reach Genie One's structured and unstructured data answers through a single entry point, with semantics constrained by Genie Ontology to reduce conflicting answers across separately configured agents.

Governance and audit logging run through Unity Gateway, which is what determines whether enterprises can approve it. Note this is a vendor announcement with no independent evaluation; the customer quote was chosen by the vendor, and real answer quality and adoption remain to be seen.

Finnish cloud provider Verda, formerly DataCrunch, announced a $189 million Series B round to expand its AI compute capacity and platform services.

Emergence Capital led the round, with participation from MUFG Innovation Partners, Supermicro, Finnish pension company Varma and others, plus two individual investors from Meta and Core Automation. The company says total raised now exceeds $450 million across equity and debt.

Valuation and round terms were not disclosed, and all figures come from the company's own announcement. Verda plans to operate more than 250 megawatts in 2027 and says it will deploy Nvidia VR200 NVL72 racks in the coming months; whether those targets land is what to watch.

Internal Meta posts seen by 404 Media confirm that the Muse feature, which calls businesses on a user's behalf, includes a human agent layer during testing, with some calls placed by trained human agents.

On September 16 Meta executives publicly promoted Muse's outbound calling. The internal posts state that Muse can hand a request to a trained human agent who places the call. Employees questioned users unknowingly handing call requests to people, and one warned of severe negative coverage.

A Meta spokesperson said this is internal dogfooding and that proper disclosures will come before any public release. How often calls are routed to humans, and under what conditions, remains undisclosed.

On September 22, OpenAI published proposed priorities and principles for third-party safety assessments, naming four priority areas: safety cases, critical safeguards, capability evaluations, and alignment evaluations covering its Preparedness risk categories.

The company says it is willing to grant assessors deep access, including technical safeguard information, chain-of-thought access, and internal deployment access, with engagements running from weeks to several months and generally not tied to specific launches.

Note the limits: this is OpenAI's own proposal, not a rule. No assessors, timeline, or consequences for non-compliance are named. Until a first actual assessment is published, the commitments cannot be verified.