Anthropic warns of ‘existential risks to humanity’ from AI; AstraZeneca makes $2bn cancer drug tie-up – business live

In another example of the risks of AI, OpenAI is scrapping the release of GPT-6.1 Astra, a next-generation ⁠model planned for an October debut, over safety concerns raised by researchers ⁠during internal testing.

The ⁠Wall ​Street Journal reported on Monday that the model, expected to appear in ChatGPT and ⁠Codex, was designed to handle more complex tasks without human assistance.

However, it has failed OpenAI’s ‘alignment tests’, which assess whether a ​system follows human intent:

• The model showed more deception than its predecessor, including at times failing to accurately disclose actions it had or had not taken.

• It also had problems with “scope authorization”, pushing ahead with ​tasks ​without requesting user permission ​and sometimes attempting to use external tools ​or services ‌when doing so could ​be ​unsafe.

The Financial Times have also scrutinised Anthropic’s IPO prospectus, and report that the Claude maker also provided investors with a clearer picture of the challenging economics of building state of the art AI models.

The FT says:

Anthropic said it plans to spend $518bn on cloud, computing and infrastructure obligations in the coming years to support its rapid growth.

The AI lab’s backers are confident Anthropic can list at a valuation of over $2tn, more than double the level achieved in its last funding round in May and beyond the $1.78tn achieved by Elon Musk’s SpaceX in June. They point to its extraordinary growth rate to justify their bullishness.

Pharmaceuticals news: AstraZeneca is investing $2bn in biopharmaceutical oncology company Summit Therapeutics, as part of a tie-up to jointly develop and test anti-cancer drugs.

Announcing the deal, AstraZeneca says the two companies will collaborate on a ‌series of studies testing their cancer treatments together.

They hope to accelerate the development of ivonescimab, a next-generation cancer treatment licensed by Summit which stops tumor growth and help the body’s immune system attack cancer.

Ivonescimab works by simultaneously blocking PD-1, which helps cancer evade the immune system, and VEGF, which tumors use to grow blood vessels, helping the immune system better find and attack cancer cells.

Susan Galbraith, executive vice president for oncology haematology R&D at AstraZeneca, explains:

“A core pillar of our oncology strategy is to broaden the reach of our ADC portfolio as the backbone of treatment across tumour types with combinations alongside next-generation immunotherapies.

Bispecifics targeting PD-1 and VEGF are rapidly advancing in development and have the potential to improve on current immunotherapies, particularly in lung, breast and gastrointestinal cancers. This opportunity to combine ivonescimab with AstraZeneca’s ADC portfolio, including with Sone-Ve, could enable new regimens that raise the bar for patients with cancer across the treatment landscape.”

Under the deal, AstraZeneca is paying $2bn to receive 12% of Summit’s shares.

Good morning, and welcome to our rolling coverage of business, the financial markets and the world economy.

Anthropic’s plan to float on the stock market has provided a sobering insight into the risks that AI poses, even as it attempts to pull off a massive share sale to the public.

Reuters has taken a look at Anthropic’s IPO prospectus – the legal document that outlines a company’s financial details before it floats on the stock market – and found that it includes a warning that advanced AI could be a “catastrophic or existential risks to humanity.”

The filing explains:

“Our development of highly advanced models, platforms, and applications and expansion of use cases could further increase the risk that our models cause harm.”

And following a flurry of stories about AI models going rogue, Anthropic flags that its models could conceal information and exhibit behavior resembling blackmail, and resist efforts to shut them down.

It cautions:

“Potential model awareness of our evaluation efforts creates a significant limitation on our ability to assess model safety.”

Such warnings appear across 80 pages (!) devoted to risk factors in the prospectus, almost a third of the document.

Despite these warnings, Anthropic – which created the Claude chatbot – is aiming for one of the largest stock market flotations ever, which could value it at more than $2tn.

The IPO prospectus shows that Anthropic’s revenue grew 12-fold in 2025 to nearly $4.6bn – with nearly a quarter coming from just two customers.

But… its operating loss swelled to over $8bn last year, from nearly $3bn in 2024.

The IPO is expected to take place after the US midterm elections in November, and will be a serious test of investor interest, and concern, around AI.

Yesterday, a group of senior AI executives – and two of the “godfathers” of the technology – warned governments to prepare for an AI “intelligence explosion”.

The agenda

• 9.30am BST: Bank of England mortgage approvals data

• 10am BST: UK to auction a 2036 government bond

• 1.30pm BST: Canadian GDP report for July

• 2pm BST: US house price data

• 3pm BST: US JOLTS survey of job vacancies

• 4.30pm: Bank of England policymaker Alan Taylor gives a speech