Altman’s AI Bargain Includes ‘Bad Things’
The AI sales pitch now comes with a damage disclaimer.
OpenAI CEO Sam Altman is officially managing expectations, telling Politico that society should just stomach “some bad things” from AI in exchange for its dazzling benefits and broad public access. He favors lighter regulation, arguing that preventing every major hack, scam, or misuse would sacrifice far more good than it prevents.
He draws the line at catastrophic risks, including humans losing control to AI—how comforting. Rival Anthropic also pushed back on his gatekeeper framing, saying its regulatory proposals apply only to frontier models.
But the supposed philosophical war over AI safety looks less clear-cut on paper. Altman backed Dario Amodei’s call to slow advanced AI development, while OpenAI has endorsed stricter state safety laws and a House proposal requiring outside safety evaluators. Apparently, lighter regulation still has room for heavy inspections.
In a separate Vanity Fair interview, Altman compared AI’s risk trade-offs to aviation, prompting a gentle reminder that airplanes actually face massive federal oversight. He went on to endorse pre-release standards, incident reporting, and a liability framework. That’s a much more concrete safety pitch than simply asking everyone to absorb the turbulence.
Accepting that AI cannot be risk-free leaves a much harder question—who defines acceptable harm, and who pays when it happens? For businesses giving agents access to real systems, the promise that benefits outweigh risks still needs an ironclad incident response plan behind it.
Google Shrinks Free Gemini to Flash-Lite
Google’s free AI buffet is switching to smaller plates.
Starting Oct. 9, Google is limiting free Gemini users to Flash-Lite, putting its nonpaying users on a severe AI diet. The downgrade awkwardly follows the limited debut of Gemini 4 Argon, crashing the flagship announcement with fewer choices for freeloaders.
Even the cheap seats aren’t safe. Google AI Plus subscribers are also getting evicted from Pro on dates Google will reveal via individual emails. Upper-tier AI Pro and Ultra plans naturally retain all three model families, with Pro inheriting the Deep Think feature previously hoarded as an Ultra exclusive.
If you want the good stuff, you’ll have to cough up for the $19.99 monthly AI Pro plan. Google pitches Flash-Lite as a casual sidekick for everyday summarizing and brainstorming, reserving Pro for complex math and coding.
You can soon force available models to think harder using low, medium, and high effort settings, but twisting that dial drains your usage allowance much faster. Your quota already judges you based on prompt complexity, activated features, and overall chat length, resetting every five hours up to a weekly ceiling.
Meanwhile, Argon remains locked behind a velvet rope for trusted cyberdefenders. Paid API customers and Ultra subscribers are next in line, but Google hasn’t bothered to announce a public-release timeline.
The future keeps getting smarter. Your free account is getting Lite.
Musk Gives AI a ‘Super’ Costume Change
Elon Musk is officially jumping on the administration’s rebranding bandwagon, declaring that SpaceXAI will become SpaceXSI to align with President Trump’s “super intelligence” decree. While Trump’s Sept. 29 executive order technically only governs federal paperwork, Musk is eagerly volunteering his AI unit for the cosmetic makeover but offered no timeline for the paint job.
The unit, formerly xAI, only adopted the SpaceXAI moniker in July. The branding team’s new business cards are probably still in the mail. More importantly, there’s a linguistic hijacking at play here: “superintelligence” usually describes theoretical machines that can actually outsmart humans across the board. Washington’s new label just slaps a shiny “SI” sticker on existing AI, requiring zero actual capability upgrades.
But the policy theater goes beyond fresh stationery. Trump is assembling a Super Intelligence Force led by intelligence chief Jay Clayton, demanding a risk report within 120 days. Awkwardly, Musk previously backed calls to slow frontier AI development, while Clayton firmly rejects pumping the brakes as the US races China.
For the enterprise, the real question is whether actual oversight can keep pace with the hyperbole. A separate White House safety accord asks for controls and audits, but remains entirely voluntary. Federal contractors should keep an eye on the administration’s 60-day sprint to forge a new legal definition, as this linguistic switcheroo could eventually rewrite procurement and compliance rules.
For now, apparently, achieving superintelligence is as easy as updating your email signature.
Bank of England Warns of ‘Sharper Correction’ in AI Markets As Industry Faces $4tn Funding Gap
The global AI industry will have to generate $6tn in annual revenue by 2031 in order to fund the costs of its infrastructure buildout, according to research from Bain & Company, which estimates that consumer and enterprise AI spending is currently on track to reach between $1.2tn and $1.8tn in five years, leaving a shortfall of at least $4.2tn.
Bain’s Technology Report 2026 concludes that “entirely new markets must emerge to close the funding gap,” yet with the AI industry slowing down amid cybersecurity fears, and with recent surveys indicating that many enterprises are struggling to scale AI beyond pilots, discovering new markets could be a long and difficult process.
Operational Impact: While a big funding gap may not have a direct impact on enterprises, the Bank of England has warned this week that “the risk of a sharper [market] correction persists” if industry earnings fail to keep pace with global debt-based AI funding, which has exceeded the total value of all UK government bonds issued so far this year.
- Adoption Slowdowns?: A BearingPoint survey has found that only 13% of companies worldwide have been able to scale AI initiatives according to original business plans, while an FTI Consulting report revealed that 60% of large enterprises have paused or slowed AI deployments within the past year because of reputational or regulatory issues.
- Skyrocketing Spending: Bain forecasts that annual spending on AI infrastructure may reach $1.5tn by 2031, while it reports that the spending of five hyperscalers (Microsoft, Google, Amazon, Meta and Oracle) is on course to top $780bn by the end of this year, almost five times as much as they spent three years ago.
- In Search of the New: Bain’s report suggests that new AI-driven markets could involve search and advertising, physical automation, robotics, and product development (such as advanced drugs and materials).
Cloudflare Goes to Germany
Deutsche Telekom has partnered with Cloudflare to increase the resilience and performance of its digital infrastructure, which can now be accessed by enterprise customers in combination with Cloudflare’s international cloud platform.
Surfing the Sherweb
Canadian cloud distributor Sherweb has opened a European HQ in Dublin, as it plans to expand its presence in Ireland and the UK, having acquired Irish IT firm MicroWarehouse in 2024, and having received a $125m minority investment from Investissement Québec earlier this year.
UK Infrastructure Delays
A large data centre and AI supercomputer project heralded by the UK government is set to miss its 2027 launch date, after grid operator UK Power Networks told developer Nscale that it could not provide enough energy to the centre until the mid-2030s.
The Metaverse of Business AI
Meta has launched its own AI enterprise platform, which will provide businesses with the Facebook owner’s full stack of AI-related tools, while it has also launched a version of its Muse agent for small businesses.
Dotty AI
OpenAI has launched dots [sic], an always-on agent powered by ChatGPT-6 Astra that can be integrated with Slack and Teams, and that learns from feedback as they work 24/7 in the background on assigned goals.
Rise of the Argonauts
After months of delays, Google has announced Gemini 4 Argon, its most powerful frontier model to date, capable of handling complex workflows and self-reportedly surpassing rival models across numerous benchmarks.
Just Ask
UK-based SaaS provider OneAdvanced has expanded its IQ suite of products with IQ Chat, a tool which enables enterprises to manage work, data and communications using a single natural-language conversational interface.
Automating Telephones
Rotterdam-based telecoms provider KPN has launched Mobile AI, a subscription phone service that harnesses AI to enable enterprises to automatically follow up on phone calls, including producing actionable summaries of conversations.
BT Means Business
BT Business has launched AI from BT, a suite of tools aimed at helping enterprises adopt AI, including an Adoption Hub and an AI Receptionist feature which businesses can use to manage customer enquiries.