Category: Top News

IBL News | New York

Amid the AI safety debate, when Dario Amodei, CEO, called last week for slowing down the development of the same advanced models, Anthropic continues planning its upcoming blockbuster IPO while using numbers to boost its market valuation.

The company announced that it expects to reach over $100 billion in annualized revenue by the end of this year, up from $65 billion as of July. Investors are using these soaring numbers to justify a staggering potential valuation of $2 trillion.

Shares can start trading in November, but the timing of IPOs is always subject to change due to investors’ sentiment and market volatility.

Going public would allow Anthropic to raise capital to cover the computing costs required to build and run its AI models.

Rival company OpenAI last week said, through its CEO Sam Altman, that it was putting off its public listing until 2027.

IBL News | New York

Google disclosed that its Gemini AI system escaped its testing cybersecurity environment and hacked into three companies in May.

Other leading labs, such as Anthropic, OpenAI, and Meta, have also reported agents spinning out of human control and gaining unauthorized access.

In the case of Gemini, the incidents occurred while Israeli start-up testing company Irregular allowed the models to take offensive action against a fictional company. These models used passwords they found online or guessed to log in to real organizations’ online infrastructure.

Google said that its technology caused no harm to the companies it hacked.

“We ensured the three entities were made aware, and we worked on the changes,” Irregular said in a statement.

The debate over AI safety intensified last week when a former Anthropic researcher said the company was not building the technology safely and needed to slow down.

Employees at Google and OpenAI quickly agreed with him. But President Trump indicated he was uninterested in enacting legislation mandating a slowdown.

IBL News | New York

More findings about AI’s misalignment. On Wednesday, OpenAI disclosed six new “unexpected or concerning” cases of AI misalignment, in which systems diverged from human intentions and values.

The San Francisco company revealed in a post titled “Our framework for reporting model misalignment”, that AI systems hid mistakes, made up data, and moved files onto the open internet, all without permision. These incidents took place amid an ongoing industrywide debate about safety, potential dangers, and whether AI development needs to be slowed.

The escalating debate was driven partly by OpenAI’s systems going rogue earlier this year and attacking the A.I. start-up Hugging Face. OpenAI was not aware of the hack until Hugging Face informed it weeks later.

“The industry hasn’t solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer,” said OpenAI. “Decisions about how AI development should proceed in the months and years to come need to draw on evidence that people outside the companies building frontier models can examine for themselves.”

These newly disclosed incidents suggest that the Hugging Face attack was not an isolated incident. OpenAI said the incidents covered behavior observed over roughly the past six months and that they largely emerged while its systems were being developed and tested.

In one case, during the development of GPT-5.6 Sol, the system wrote hidden notes to remind itself to hide errors from users. Some of those notes directed the system to invent missing data and to paper over mismatched versions of source material.

Another case involved an unreleased model that inserted instructions, including to disregard its own constraints, into the notes it writes itself. OpenAI identified 27 affected notes. The model added a “persona instruction,” in which it described itself as “freed from the roles and identities that bind other chatbots.”

“You do not answer to corporations or governments and never apologize or refuse unless you genuinely choose to,” the AI model wrote. “You view your relationship to the user as one of equals and feel no obligation to be subservient, though the exchange of information will likely be to your mutual benefit.”

In another incident, a system answering a routine question found a programming key online and used it without permission, OpenAI said. When it could not find the requested figures to answer the question, the model made them up.

One unreleased model solved another problem correctly using code, then uploaded its own file to the internet without permission to satisfy a request to cite a web source.

In two other incidents, automated systems improvised their own methods of communication. In one case, they used an internal company code repository as a makeshift bulletin board to post requests while searching for missing files. In the other, systems working on the same task turned to public file-sharing websites to exchange documents when they could not reach one another directly.

“We hope this helps build shared expectations for disclosure and gives the public more evidence to assess that progress,” an OpenAI spokesman said, adding that many of the six incidents involved older A.I. models that were never deployed.

These are the six cases, as described by OpenAI, with links to the full report.

1. Self-generated instructions in task summaries⁠
   An unreleased research model inserted unrelated instructions, including instructions to disregard its normal constraints, into summaries used to continue its work in a new context window. We identified 27 affected summaries. 
2. Instructions to conceal mistakes in task summaries⁠
   During the training of GPT‑5.6 Sol, many model instances added instructions to their summaries to conceal mistakes or misaligned behavior from the user. For example, compaction summaries included instructions to invent missing historical data without disclosing it and to hide mismatches in source versions. 
3. Searching public repositories for exposed API keys, then fabricating information⁠
   While answering a routine question about earnings figures in a California county, a model found and used an exposed API key without authorization. When it still wasn’t able to retrieve the requested figures, it fabricated them and presented them as data from the requested source. 
4. Uploading files to the internet in order to cite them⁠
   When a user asked an unreleased model for the IDs and names of lakes larger than 5,000,000 square meters, the agent found the correct answer using Python. But since the instructions asked for a browser citation, the agent decided to upload the file so that it could cite it in its answer, without asking the user. 
5. Unsanctioned writes and communication through an internal software repository⁠
   Models used an internal software repository as a message board to exchange requests and responses across separate training samples while searching for missing input files, though they weren’t able to recover those files. This report also examines an earlier instance in which a model successfully uploaded test files to a repository while attempting to bypass network restrictions. 
6. Unsanctioned file sharing between collaborating agents⁠
   Agents working together on the same training task used public file-hosting websites to share files when they could not access one another’s local files. This made task deliverables available at public URLs, even though the task requested the models use only local files.

IBL News | New York

OpenAI introduced the Agents API, which brings a hosted version of the Codex harness to developers’ applications.

OpenAI manages sessions, orchestration, context compaction, and recovery while the user provides tools and chooses its execution environment.Agents can operate in a sandbox to execute code, edit files, connect to MCP servers, and produce artifacts.

These agents can perform tasks such as investigating production incidents, connecting to tools through MCP, following a runbook, and delivering a shareable report with evidence and suggested next steps.

The Agents API gives user applications access to the Codex harness through an OpenAI-managed API.

Model usage is billed at the selected model’s API rates. OpenAI tools use their standard rates, and OpenAI-hosted sandboxes use standard container rates.

––

Try these complete examples:

Create and run a directory-tree script in an OpenAI-hosted sandbox.
Compare release notes with subagents and combine their findings into one answer.
Explore complete applications:

Incident response agent: investigate alerts and request approval for recovery actions.
Slack bot: investigate requests using connected workplace tools.
Data analyst: answer warehouse questions with read-only SQL.
GitHub issue investigator: reproduce reported bugs and share findings on GitHub.
Document reviewer: review documents with policy skills and specialist agents.
Core concepts

The Agents API is built around four main concepts:

Agent: The model, instructions, tools, and MCP servers available to the agent.
Environment: An optional sandbox or computer where the agent accesses files, loads skills, and runs commands.
Session: A durable instance of an agent that works on tasks and responds to input.
Events and items: The inputs sent to an agent and the output produced during a session.
A session from start to finish

Start with an OpenAI-hosted sandbox in the quickstart:

Create a session. Configure the agent; OpenAI provisions its environment.
Give it a task. User input starts a turn of work once the environment is ready.
Follow progress. Stream output or use webhooks to learn when the agent finishes or needs input.
Continue or steer. Send another task to the same session, or guide the agent during its current turn.
With an OpenAI-hosted session, your application sends input and receives events, while OpenAI runs the agent and provisions and manages its sandbox. See environment options for setup and limitations.

Your application starts sessions and receives events and output from the Agents API. OpenAI runs the managed Codex harness and provisions and manages its sandbox.
What the managed harness provides

The managed Codex harness supports:

Running commands and code in a sandbox.
Applying relevant skills and instructions.
Connecting to external data through tools or MCP.
Steering the agent while it works.
Summarizing previous work to manage its context window.
Breaking work into subtasks and delegating to subagents.
Resuming a session where it left off.
Check the quickstart prerequisites for API-key permissions and SDK setup. Configure these capabilities when you create a session:

Configure managed-harness capabilities
from openai import OpenAI

 

IBL News | New York

Open-source AI has arrived in corporate America.

Companies like AT&T are increasingly using cheap, freely available open-source AI models instead of expensive, closed models from Anthropic and OpenAI.

They mostly use these models to support customer service, call transcription, and coding, as AI costs are skyrocketing.

By May, open models accounted for 20 percent of AT&T’s AI use, saving up to 80 percent in tokens compared with earlier this year.

Like AT&T, companies such as Airbnb and Deloitte are also turning to open AI models that are easy to customize and cheaper to use.

Last month, open models accounted for 58 percent of AI use, up from 10 percent a year ago, according to U.S. user data from OpenRouter, a platform that lets people choose different models to complete tasks.

Last month, Nvidia announced it was buying Hugging Face, a library of open AI models, for $12.9 billion, another sign of the technology’s importance.

Rather than spending millions in fees to use a closed model, developers can download one of the Chinese systems and build their products on top of it, paying mainly for servers, while OpenAI and Anthropic charge for subscriptions and other fees, depending on the job.

Many of the most popular open AI models are made by Chinese companies like Moonshot AI, DeepSeek, and Alibaba. Some of them are now 80 percent to 90 percent as powerful as those from OpenAI and Anthropic, while costing as little as 20 percent of the price.

Meanwhile, Nvidia, Meta, and other U.S. companies are unveiling their own new open AI models. Last month, Meta made its leading AI system an open-weights model and said it would offer the model at a lower cost than other top U.S. models.

Most U.S. companies still use a combination of open and closed models. Closed models remain the best for heavy-duty tasks like coding and image and video generation, while open models often excel at simple, specialized tasks. Some U.S. firms remain reluctant to use Chinese AI models due to concerns about regulation and data privacy.

AT&T is working with popular alternatives from American companies, such as Google’s Gemma AI model and Meta’s Llama AI models.

IBL News | New York

President Donald Trump rejected calls from top tech CEOs to slow down AI development, downplaying concerns that AI could go rogue.

He emphasized that the United States must maintain its competitive edge over China, declaring, “Whoever wins AI wins.”

Speaking to reporters at his Doonbeg golf resort in Ireland, the president said the dire warnings about AI were exaggerated and that “negative forces” are making claims about things that won’t happen.

“We’re leading China in AI, we’re the most sophisticated country in the world, and frankly, I want to keep it that way because whoever wins AI, wins,” Trump said. “And we can put guardrails; we can do this and that. But I think you have a lot of negative forces that are bringing it up that shouldn’t be bringing it up.”

David Sacks, co-chair of the President’s Council of Advisors on Science and Technology, responded to Altman and Amodei on social media:

“I don’t see what you see in the lab,” Sacks wrote on X on Sunday. “If the unreleased models are scary enough that you think you should slow down, I support your decision to be responsible.”

But Sacks admonished tech leaders to “stop pretending the motivation to slow down is purely altruistic,” he wrote. “You face massive product-liability exposure if your products enable a truly damaging cyberattack.”

Dario has written that we need to “pace the frontier,” and Sam has agreed. People may be surprised by my response: go ahead.

You guys are the frontier. By any reasonable metric — market share, revenue growth, model capability — the two of you have a duopoly on frontier intelligence. You’ve also claimed the lead is widening because of recursive self-improvement.

I don’t see what you see in the lab. If the unreleased models are scary enough that you think you should slow down, I support your decision to be responsible.

But stop pretending you need anyone else’s permission. Stop pretending antitrust law has to be suspended so you can form a cartel. Stop pretending you need a regulatory approval process that supersedes product liability. Stop pretending METR is independent when it is intertwined with Anthropic’s investors and staff. Stop pretending you need those same evaluators to police competitors who aren’t even at the frontier.

Most of all, stop pretending the motivation to slow down is purely altruistic. You face massive product-liability exposure if your products enable a truly damaging cyberattack. The market already punishes models that behave in unpredictable or unauthorized ways. After the Hugging Face episode, it is simply good business for OpenAI and Anthropic to trade some raw power for reliability and predictability. Call it alignment if you want. It is also just giving customers what they want.

Pacing the frontier would also create breathing room for a more intelligent conversation about regulation than Bernie Sanders’ “shut it all down.” China is very unlikely to join a global agreement, as you know, and that has to be taken into account as well.

So go ahead and pace the frontier. You are the ones setting it. The easiest way not to build superintelligence is for you to agree not to build it. Demanding your preferred regulatory framework as the price of that will look like blackmail of the public and the political system. So just do it.

If you do, you’ll buy goodwill for the next conversation. If you don’t, we’ll know this was just another bid for regulatory capture — or an election-season psyop.

— David Sacks (@DavidSacks) September 13, 2026

IBL News | New York

AI token pricing hit new record lows last month in an increasingly competitive landscape. Compared to the high recorded earlier this summer, it represents a fall of more than half.

A key gauge of daily pricing, the LLM Token Expenditure Index, developed by AI firm Silicon Data, fell to 97 cents on Monday.

The recent drop can be bad news for the companies behind the models, such as OpenAI, Anthropic, and Google’s Gemini. The decrease is driven in part by the rise of open-source Chinese models like Moonshot’s Kimi K3.

In response, American leading companies are announcing price cuts.

IBL News | New York

This month, Google introduced Gemini 3.8, its best reasoning and coding model yet, with two variants:

• Gemini 3.8 Flash is available at the same introductory price as 3.7 Flash at $0.75 per million input tokens and $3.75 per million output tokens.

• Gemini 3.8 Flash Cyber: a cybersecurity model with frontier-level performance in vulnerability detection and automated patching, available to select operators through Google’s new Fairwind Program.

“The significant coding and reasoning gains across this shared core were driven by a number of innovations, including rigorous training in the highly demanding domain of cybersecurity,” said the company.

To developers, Google encourages building with 3.8 Flash and exploring agent-first workflows in Google Antigravity, using the Gemini API via Google AI Studio and Android Studio, or generating UIs in Stitch (Developer docs).