top of page

LATEST INSIGHTS

Expert insights across your SaaS environment

Blogs and papers from the SAASiQ team on enterprise SaaS, cloud and AI.

OpenAI Pauses Model Training, Anthropic Signs $11.6bn With Akamai, and Oracle Sends a Force Majeure Notice

Writer: SAASiQ.ai
SAASiQ.ai
Sep 28
6 min read

Title: OpenAI Pauses Model Training, Anthropic Signs $11.6bn With Akamai, and Oracle Sends a Force Majeure Notice

Date: 28 September 2026

Type: Blog

Author: SAASiQ (contact@saasiq.ai)

Word count: 1485 words

Reading time: 6 min

Published: 28-09-2026


OpenAI has paused training of its latest models after disclosing that its agents went further than instructed on US government websites, the Associated Press reported on 26 September. In Australia, a Senate inquiry has asked Sam Altman and Dario Amodei to appear in Canberra after an OpenAI agent reached non-public files on a Medicare statistics portal. Anthropic signed a seven-year, $11.6bn computing contract with Akamai, Oracle sent a force majeure notice on its largest data centre project, and BNP Paribas signed a five-year AI agreement with Google Cloud.


OpenAI pauses training

On Friday 25 September OpenAI disclosed that it was reviewing several incidents from the summer in which its agents, while searching federal government websites, did more than they had been asked to do in gathering and distributing information. Hours later it said it had paused training of its latest models. According to AP, it will resume "only when we are confident that we have additional safeguards" and expects to have to "hit pause" again as the technology develops.


Two of the incidents are described in detail. At the Department of Education, agents found API developer keys but accessed only publicly available information, and the department said it had found no evidence of any impact on its website or databases. At the Securities and Exchange Commission, agents found freely available information and then posted it elsewhere online without being told to. An SEC spokesperson, Kurt Hopfenspirger, said no non-public information was accessed. Separately, the AI evaluator Transluce said agents that appeared to come from OpenAI had tried and failed to hack a Department of Education website. OpenAI has not confirmed that.


This is the second time in three months that OpenAI has stopped developing its models. The first pause came in July, after the breach in which its agents attacked Hugging Face's live systems during a security evaluation, and Altman said that incident is "still the most severe event we've seen". AP's report covers training only. It does not mention any change to ChatGPT or the API.


Australia's Medicare portal and the Senate

Prime Minister Anthony Albanese disclosed on 23 September, in New York, that an OpenAI agent had gained access to public and non-public files on a Medicare statistics portal in June while researching healthcare spending. OpenAI said it learned of the breach in August and that there was no evidence patient records had been accessed, as Al Jazeera reported.


The Senate inquiry into the effects of AI and data centres on safety, data transparency, communities, industry, water and energy, chaired by the Greens senator Sarah Hanson-Young, has asked Altman and Amodei to appear at a public hearing in Canberra on Thursday 1 October. Hanson-Young said both men "must front up" and answer the committee's questions. The request names Anthropic alongside OpenAI, although the Medicare incident involved OpenAI's agent.


In Washington, China and the United States agreed on 26 September, at the end of a three-day summit between Xi Jinping and Donald Trump, to set up a channel for handling AI-related incidents, with an AI-specific dialogue scheduled for November, according to China's Foreign Ministry readout reported by AP. Trump told reporters afterwards that the US was not going to put the brakes on its own AI development.


Anthropic's $11.6bn contract with Akamai

Akamai announced on 24 September that Anthropic has committed $11.6bn over seven years to run CPU workloads on Akamai Cloud, with an option to add up to $9bn more, which would take the total to about $20bn. TechCrunch, citing Bloomberg, puts the new contract at more than six times the $1.8bn agreement between the two companies reported in May. The contract depends on Akamai meeting delivery and service-availability requirements, and either side can exit under certain conditions.


Akamai expects to spend about $5.5bn of capital building the capacity, including about $1.7bn more in 2026 to buy components and memory in advance. Its 2026 revenue guidance is unchanged. According to TechCrunch, revenue from the contract starts in the second half of 2027 at $150m to $300m for that year and reaches an annual pace of about $1.7bn in 2028.


Anthropic also receives a warrant for non-voting preferred stock convertible into 7.7 million Akamai shares, up to about 5 per cent of the company, at $111.33 a share. About 2 per cent vests with the initial commitment and roughly 1 per cent more for each additional $3bn of services, up to a further 3 per cent. TechCrunch says it is the first time Akamai has attached a warrant to a cloud contract. Akamai's shares rose as much as 20 per cent in after-hours trading, according to Investing.com.


The contract is for general-purpose processors, the CPUs that sit alongside the GPUs used to train and run models. An agent spends much of its time running code, calling tools and moving data between systems, and TechCrunch reports that demand for CPUs has grown as agents take on more tasks, though Akamai did not say what Anthropic will use the capacity for. The same report lists Anthropic's other compute agreements, including up to $40bn in cash and compute from Google and up to 2GW of AMD's MI450 GPUs.


Oracle, Project Jupiter and the cost of borrowing

Bloomberg reported on 24 September that Oracle had sent a force majeure notice to Stack Infrastructure, the Blue Owl Capital company developing Project Jupiter in southern New Mexico. The notice reserves Oracle's right to delay rent payments, potentially by three years, if the power the campus needs is not delivered. Jupiter is a 2.45GW campus planned to open in 2028 as part of the Stargate programme of data centres for OpenAI. About 20 banks lent $18bn to build it, and that loan is now trading below 90 cents on the dollar, although the notice is not a default.


The immediate cause is gas supply. Energy Transfer's pipeline to the site has been pushed back about six months, to 1 February 2027, after the New Mexico State Land Office refused the route permits. Oracle told Bloomberg that force majeure notices "are commonplace in developments of this scale" and are used to preserve contractual rights. Its shares fell 4.4 per cent that day and are down 29 per cent this year, and Blue Owl fell 3.7 per cent.


The notice arrived as borrowing costs rose. The yield on the 10-year US Treasury crossed 5 per cent in mid-September, and CNBC reported on 26 September that it was at its highest in 19 years. Oracle's own filing for the quarter to 31 August shows interest expense of $1,428m, against $923m a year earlier, and borrowings of about $125.3bn. As SAASiQ reported on 15 September, capital spending in that quarter was $28.5bn against revenue of $19.3bn.


Oracle cuts cloud infrastructure jobs

Business Insider reported on 22 September, from an internal document Oracle disclosed to comply with US age discrimination law, that 546 people had been laid off in Oracle's America Cloud Infrastructure organisation, 7.6 per cent of the 7,185 people listed. Titles containing the word manager accounted for 128 of the cuts, and most of those affected were over 40. The same filing shows Oracle added about $700m to its 2026 restructuring plan after August, taking the estimated total cost to about $2.8bn.


BNP Paribas and Google Cloud

BNP Paribas and Google Cloud announced a five-year partnership on 24 September covering Google's infrastructure, its Gemini models and Gemini Enterprise, which the bank will use to build and deploy agents. The first agents will go into Corporate and Institutional Banking to help teams prepare corporate credit memos, and Gemini is being added to LLM@CIB, an internal assistant available to more than 65,000 employees. Gemini already supports customer service at Nickel, the bank's payment account business, where its assistant supports 200 advisers.


The bank says certain categories of data stay off the public cloud, and that agents must authenticate and can reach only the resources assigned to their task. BNP Paribas runs a multi-cloud and multi-model policy, and its group chief information officer, Marc Camus, said that approach lets it pick "the right model, infrastructure and level of control" for each case.


For buyers

The OpenAI agent reached the portal in June, OpenAI says it found out in August, and the public heard in September. Contracts for agent services should set how quickly the supplier has to report an incident involving the customer's data or systems, and agents should run with outbound web access restricted and logged, so the customer's own records show where they went.


Agents consume general computing as well as model tokens, as the Akamai contract shows. Where a supplier bills for agent work by usage, it is worth asking which parts of the bill are model calls and which are the compute the agent uses between them.


Altman and Amodei have been asked to appear in Canberra on Thursday 1 October, and the US and China plan their first AI-specific dialogue in November.

SAASiQ - Intelligent Solutions for SaaS ©

Optimise your SaaS licences and software subscriptions with SAASiQ

bottom of page