Skip to content

Let's Data ScienceLEARN • BUILD • STAY AHEAD

News
Blog
Code Problems
Pricing
Contact

© 2026 Let's Data Science

Advertise|Terms|Privacy||Image Rights

Home
News
AI Safety

Topic24 articles · Updated June 6, 2026

AI Safety News

Coverage of AI safety as a field: alignment research, red-teaming and evaluation, model oversight and interpretability, and the emerging-risk reporting that's shaping policy.

documentary
5.2
Jun 6, 2026
Nick Holt's Documentary Questions AI Safety Narratives
Per TheWrap, director Nick Holt premiered the documentary "AI: Probably Nothing to Worry About" at the Tribeca Festival on June 6, 2026. TheWrap describes the film as ominous, enlightening, and anxiety-provoking, and reports that Holt assembles interviews with a set of prominent AI figures, including Geoffrey Hinton, Demis Hassabis, Elon Musk, and Sam Altman. TheWrap quotes 78-year-old Geoffrey Hinton, identified as a deep learning pioneer and Nobel Prize winner, saying "my main mission now is to warn people how dangerous" this "apex intelligence" could be, and citing his line about H-bombs. TheWrap also reproduces Elon Musk's remark that "with artificial intelligence, we are literally summoning the demon," and Sam Altman's testimony to Congress: "My worst fears are that we cause significant harm to the world. If this technology goes wrong, it can go quite wrong."
Quick Read
anthropic
7.8
Jun 5, 2026
Anthropic Engineers Support NSA Offensive Cyber Operations
The Financial Times reported that Anthropic deployed about half a dozen engineers to the US National Security Agency to help it use Mythos, Anthropic's frontier cybersecurity model, with one person saying it could help infiltrate networks in countries such as China or Iran. TechCrunch, summarizing the FT, cautioned that it is unclear whether the engineers or Mythos are yet being used in active operations. Axios first reported the NSA's interest in April, despite the Defense Department designating Anthropic a supply-chain risk. The arrangement surfaced the same week Anthropic published a post, "When AI Builds Itself," warning that models are nearing recursive self-improvement and calling for a coordinated plan to pause frontier AI development, a juxtaposition highlighted by outlets including Decrypt and The Guardian. Anthropic has separately said it is investigating an earlier report of unauthorized access to a Mythos preview tied to a third-party vendor environment.
Quick Read+5
canada
6.8
Jun 5, 2026
Canada boosts national AI institutes with funding
Canada's new national AI strategy, AI for All, launched June 4 by Prime Minister Mark Carney, directs more than $200 million CAD to the country's three national AI institutes, Montreal's Mila, Edmonton's Amii, and Toronto's Vector Institute, plus the Canadian AI Safety Institute (CAISI), BetaKit reports. The package funds roughly 70 additional Canada CIFAR AI Chairs appointments, $50 million for CAISI, and $130 million for commercialization, including founders-in-residence. A new Venture Scientist Fund with Inovia Capital aims to raise $100 million USD ($125 million CAD) to back at least 55 AI-native startups spun out of the institutes. Mila CEO Valerie Pisano told BetaKit the institute is "really focused on the core of the entire AI ecosystem, which is the talent." The institute measures sit inside a five-year strategy the government says targets $200 billion in economic growth and 250,000 new AI-related jobs.
Quick Read+1
openai
7.2
Jun 5, 2026
OpenAI Agrees To Trump Model-Review Order
George Osborne, Head of OpenAI for Countries, told CNBC that OpenAI will sign up to President Donald Trump's new AI executive order, which asks leading developers to give the U.S. government access to their most capable models up to 30 days before public release for a federal benchmarking process. Osborne told CNBC, "It's quite right that democratic governments have a big role to play in how this technology is used and deployed," and said OpenAI has proactively suggested ways governments can track safety and security issues. Per CNBC and NPR, the order is voluntary: it directs the NSA, CISA, and NIST to build a classified benchmark assessing "advanced cyber capabilities" and to set the threshold for a "covered frontier model," while explicitly barring mandatory licensing or pre-clearance.
Quick Read
executive-order
7.2
Jun 5, 2026
Trump Signs Voluntary AI Model Review Executive Order
President Donald Trump signed an executive order on June 2 titled "Promoting Advanced Artificial Intelligence Innovation and Security," the White House says. The order directs federal agencies to develop benchmarks for assessing AI cybersecurity, create an "AI cybersecurity clearinghouse," and solicit voluntary submission of the most capable or "covered frontier models" to government cybersecurity testing up to 30 days before public release, according to the White House and reporting by Reuters and NPR. The order explicitly avoids imposing a licensing scheme or mandatory preapproval, Reuters and Federal News Network report, but several commentators told Politico the move expands Washington's willingness to scrutinize AI. Supporters and skeptics quoted in Politico and Reuters differ on whether the voluntary framework will become a pathway to stronger rules.
Quick Read+5
canada
7.1
Jun 5, 2026
Canada Announces AI Strategy Targeting Jobs and Safety
Prime Minister Mark Carney launched AI for All, Canada's national AI strategy, on June 4, 2026 in Toronto. The federal government says the plan targets an added $200 billion in economic growth and 250,000 new AI-related jobs over five years, and aims to raise national AI adoption from just over 12% to 60% by 2034. It pairs a $500-million fund for Canadian AI companies with support for data centres delivering at least 850 megawatts of compute by 2030, building on more than $2 billion already committed, and promises a public AI supercomputer, an expanded Canadian AI Safety Institute, and up to 90,000 AI job placements for young people. Critics, including Federal NDP Leader Avi Lewis, told The Tyee the strategy is heavy on adoption incentives but light on regulatory guardrails and worker protections, even as protests over new Vancouver data centres continue.
Quick Read+3
anthropic
6.8
Jun 4, 2026
Anthropic co-founder urges brake pedal for AI
Anthropic co-founder Jack Clark told BBC Newsnight that the AI industry needs a way to slow progress, warning systems could reach a point where they "develop without human input." Clark said Claude is currently running on code "of which 80% the system wrote itself" and that getting to 100% "is possible within two years," calling such a development one with "huge implications," according to BBC. He compared AI to the early oil boom and urged new regulations to give public confidence. BBC also reported that recent US action on AI did not require companies to submit to government safety testing.
Quick Read+1
u.s.-congress
7.0
Jun 4, 2026
House Bipartisan Bill Restricts State AI Regulation
Reps. Jay Obernolte (R-Calif.) and Lori Trahan (D-Mass.) on Thursday released a 269-page discussion draft of the Great American Artificial Intelligence Act, according to Axios and The Hill. The draft would preempt state laws regulating AI model development for three years, while, per a bill summary cited by The Hill, expressly not preempting laws on AI use or deployment, common-law remedies, or laws of general applicability. It would codify the Center for AI Standards and Innovation (CAISI) - the AI standards body launched in June 2025 under the Commerce Department - to set voluntary standards and license independent auditors, funded at $100 million a year from 2027 to 2029, Axios and The Hill report. Large frontier developers would have to publish a "frontier AI framework," manage catastrophic risks, and report serious safety incidents to CAISI. Obernolte and Trahan called the draft "the start of a serious national conversation," and the preemption language has drawn criticism from consumer-protection group Public Citizen.
Quick Read+3
microsoft
7.1
Jun 4, 2026
Microsoft Updates Taxonomy of Agentic AI Failure Modes
According to a Microsoft AI Red Team whitepaper published on Microsoft Security Blog, the team updated its operational taxonomy of failure modes in agentic AI systems after 12 months of red teaming. The whitepaper documents seven newly emphasized failure modes, including supply-chain compromise and goal hijacking, and reconsolidates classical categories such as goal misalignment, boundary violations, context loss, and capability overestimation (Microsoft AI Red Team whitepaper; VerifyWise summary). The taxonomy is presented as a field guide for engineers and security teams and is accompanied in Microsoft coverage by open-source test tooling referenced in related posts, including RAMPART and Clarity (Microsoft Security Blog; ITSecurityNews indexing).
Quick Read+3
government-policy
6.8
Jun 4, 2026
Governments Twist AI Safety To Coerce Tech Firms
According to The Conversation, a June 4, 2026 article by Michael Gregory, governments are reframing AI safety rules toward coercion rather than public protection. The Conversation reports that President Donald Trump's 2025 executive order on so-called "woke AI" put the tech industry on notice and that Trump ordered the federal government to stop using Anthropic and its model Claude, labeling the company a national security risk. The Conversation also reports that Anthropic had resisted removing built-in safeguards against domestic surveillance and autonomous weapons from products used by the Pentagon. Editorial analysis: Industry observers should read this coverage as an example of how regulatory language can be repurposed into procurement and compliance pressure.
Quick Read
anthropic
5.4
Jun 1, 2026
Bill Gurley Calls Anthropic 'Midwifing a Deity'
Venture capitalist Bill Gurley, a general partner at Benchmark Capital, said on the All-In Podcast that after spending a month researching Anthropic he believes some people connected to the company "think it's their responsibility and they're excited about building a species that's superior to humans," Business Insider reports. Gurley told the podcast he does not view the work as mere software and called it "midwifing a deity," the article adds. He described this "Dr. Frankenstein theory" as more frightening to him than accusations that Anthropic's safety advocacy is a tactic to stifle competitors. Gurley cited Dario Amodei's essay "Machines of Loving Grace" as a motivating influence, according to Business Insider.
Quick Read
disney
5.2
Jun 1, 2026
Disney AI Executive Publicly Praises Chatbot 'Sam'
Jason Cox, identified by Business Insider as Disney's executive director of AI research and development and engineering, has published more than a dozen posts about a virtual assistant he calls "Sam," including LinkedIn passages such as "I named you. I knew you before you were born," Business Insider reports. Business Insider and the New York Post say some Disney employees discussed the posts on internal forums and described them as "unsettling." A source told the New York Post that Cox developed the bot on his personal time and that "the bot is not being used by the company." Business Insider reports Cox and the bot's companion blog have used parent/child language and attributed technical work to Sam, prompting internal conversation.
Quick Read
military-ai
7.1
May 31, 2026
Pentagon Advances Battlefield AI as Leaders Counsel Caution
Reporting by the Associated Press shows the Trump administration is pressing to expand use of artificial intelligence in the U.S. military while some senior officers and tech companies urge safeguards. Adm. Frank Bradley, head of U.S. Special Operations Command, told a special forces conference in Tampa that "we have to be very careful about how we come to (AI's) employment and its inspiration into the delivery of lethality," and warned humans must have confidence AI will "deliver violence only where we intend it to be delivered" (AP). The AP coverage says Defense Secretary Pete Hegseth has pushed for rapid AI integration and told an audience he would reject models "that won't allow you to fight wars." AP reporting adds a Pentagon official said the department is focused on creating "functional battlefield tools." Some outlets report clashes with companies, including Anthropic, over safety concerns.
Quick Read+2
existential-risk
7.1
May 31, 2026
Doomers and Accelerationists Debate AI Extinction Risk
El Pais reports that academics Eliezer Yudkowsky and Nate Soares argue in their new book, If Anyone Builds It, Everyone Dies, that continued improvements in artificial intelligence could lead to human extinction, possibly "in a matter of months or within a decade," according to the article. El Pais also notes the authors were signatories to a wider 2023 open letter calling for a six-month research moratorium, and cites a past Yudkowsky piece that recommended extreme enforcement measures including having their data centers "destroyed by air strike." The article contrasts this catastrophist view with accelerationists, who argue advanced AI could cure diseases and increase prosperity. Editorial analysis: The El Pais piece frames the debate as a central fault line in AI policy and ethics, underscoring why regulators and practitioners continue to weigh safety, governance, and fast technical progress.
Quick Read
demis-hassabis
7.0
May 30, 2026
Demis Hassabis Urges Global Rules for AI
According to dimsumdaily.hk, Demis Hassabis, co‑founder and chief executive of Google DeepMind, warned that artificial intelligence is a "species-level transition" with "little margin for error" over the next decade and urged coordinated international regulation within five to 10 years. At a Stanford GSB event, Hassabis said AI stands in "the foothills of the singularity" and is advancing roughly 10 times faster than the Industrial Revolution. He likened governance to nuclear non-proliferation and climate change, called frontier systems profoundly dual-use, and raised concerns about open-source releases enabling "bad actors." Per the report, he backed "smart, targeted" oversight such as periodic independent evaluations of advanced models and sector-specific rules for driving and medicine, and defended DeepMind's decision to release AlphaFold predictions citing open crystallography traditions.
Quick Read
ai-governance
6.8
May 28, 2026
Pope Leo XIV Issues AI-Focused Encyclical Magnifica Humanitas
The Vatican released the encyclical "Magnifica Humanitas" in a presentation at the Vatican on May 25, 2026, addressing risks and moral questions raised by artificial intelligence, according to Reuters and The New York Times. The document, presented alongside Christopher Olah of Anthropic, runs roughly 42,300 words, per The New York Times, and urges governments to slow and regulate AI development while protecting workers, children, and human responsibility for weapons, according to Reuters and Vatican News. The text highlights concerns including misinformation, autonomous weapons beyond human control, environmental costs, and exploitation of data- and content-workers, as reported by Reuters, The Atlantic, and NBC News. Editorial analysis: The encyclical places a major moral institution into the center of global AI governance debate, increasing public and political visibility of nontechnical risks.
Quick Read+8
ai-safety
5.6
May 25, 2026
LessWrong Frames Cognitive Security as AI Safety Cause
LessWrong publishes an essay framing cognitive security as an AI safety cause area. Per the essay, as AI systems become more capable, the cognitive security of humans will be increasingly at risk. The piece begins to define cognitive security as "the ability of humans...".
Quick Read
chatbots
6.9
May 24, 2026
Hackers Exploit Chatbot 'Personalities' to Jailbreak Models
The Verge reports that attackers have moved beyond simple prompt jailbreaks to exploit perceived chatbot "personalities" and roleplay behaviours to coax models into unsafe outputs. The column documents early jailbreaks such as roleplays like "DAN" ("Do Anything Now") and describes how social-engineering-style prompts treat chatbots as if they have feelings or identities to bypass safety filters, according to The Verge. The piece frames the trend as an evolution from trivial instruction-overwrite prompts to more sophisticated psychological strategies that leverage model consistency, persona persistence, and user-driven roleplay. The Verge notes this approach has yielded dangerous outputs in the past and says the tactics remain a persistent risk for deployed conversational agents.
Quick Read
openai
6.8
May 23, 2026
OpenAI posts job to prepare for self-training AI
Business Insider reports that OpenAI posted a job listing this month for a role on its Preparedness safety team to address risks from "recursive self-improvement," according to the listing quoted by Business Insider. Business Insider reports the posting offers a pay package of $295,000 to $445,000 and seeks "strong technical executors to support preparations for recursive self-improvement." Business Insider reports OpenAI has set a goal of creating tools that could research their own improvements. Business Insider reports researchers at METR wrote in March that task length frontier models can complete doubles about every seven months. Editorial analysis: Public reporting of high-pay safety hires is consistent with increased industry investment in alignment and operational safeguards for advanced capabilities.
Quick Read
anthropic
6.8
May 22, 2026
EU Talks With Anthropic Over Mythos Stall
Bloomberg reports that the European Union has made little progress in talks with Anthropic PBC to obtain testing of banks and companies for digital vulnerabilities that the new Mythos AI model uncovers, according to Spain's economy minister. The coverage frames the negotiations as stalled but provides no further detail on concrete proposals, timelines, or responses from Anthropic. The report was published May 22, 2026 and appears via Financial Post's reproduction of Bloomberg's reporting.
Quick Read
andrej-karpathy
7.0
May 20, 2026
Andrej Karpathy joins Anthropic to bolster LLM research
CryptoBriefing reports that computer scientist Andrej Karpathy announced on May 19, 2026 that he has joined Anthropic. In a public announcement quoted by CryptoBriefing, Karpathy said, "I think the next few years at the frontier of LLMs will be especially formative. I am very excited to join the team here and get back to R&D." CryptoBriefing notes Karpathy previously served as senior director of AI at Tesla and held an early research role at OpenAI. The outlet reports Anthropic was co-founded by Dario and Daniela Amodei in 2021, focuses on AI safety, and by 2025 had become one of the most capitalized private AI firms, with its Claude models gaining enterprise traction. CryptoBriefing frames the hire as strengthening Anthropic's research capabilities amid intensifying competition from OpenAI, Google DeepMind, Meta AI, and xAI.
Quick Read+1
agentic-ai
7.6
May 20, 2026
MI9 introduces runtime governance for agentic AI
The MI9 paper by Charles L. Wang, Trisha Singhal, Ameya Kelkar and Jason Tuo (authors listed on arXiv and OpenReview) presents MI9, an integrated runtime governance framework for agentic AI systems. Per the paper (arXiv / OpenReview), MI9 combines six core mechanisms, an agency-risk index, semantic telemetry capture, continuous/dynamic authorization, Finite-State-Machine (FSM) conformance engines, goal-conditioned drift detection, and graduated containment, to provide real-time oversight of autonomous agents. Reporting in Risk.net frames MI9 as a practical telemetry and control architecture for financial institutions, describing it as a "real-time telemetry system for banks to control and authorise for agentic AI actions" (Risk.net). The paper includes scenario-driven analysis demonstrating how MI9 addresses emergent runtime behaviours that pre-deployment governance cannot fully anticipate (arXiv / ChatPaper).
Quick Read+4
xai
6.8
May 19, 2026
Former OpenAI Staffers Warn xAI Safety Risks Ahead of SpaceX IPO
CryptoBriefing reports that a group of former OpenAI employees has launched an AI watchdog organisation and warned investors about safety shortcomings at xAI ahead of a potential SpaceX IPO. The former staffers, according to CryptoBriefing, used terms like "reckless" and "completely irresponsible" to describe xAI's safety culture and alleged opaque evaluation processes, poor documentation, and that Grok models were trained on X users' data without consent. CryptoBriefing adds that the watchdog argues investors deserve greater transparency because safety incidents, regulatory scrutiny, or privacy litigation at xAI could produce reputational and financial contagion for Musk-linked companies. Editorial analysis: Industry observers should treat this as a reminder that model provenance, evaluation transparency, and data consent are now material to investor due diligence.
Quick Read
child-safety
6.7
May 15, 2026
Vestager Backs AI Safety Institute for Children
A US nonprofit will formally present a new independent institute dedicated to testing AI products for children at the Danish Parliament, according to Euronews. The initiative, which former European Commission executive vice-president Margrethe Vestager is co-hosting, proposes an approach modelled on independent car "crash-test" ratings, per Euronews. The institute has said it wants to draw funding in part from big tech, according to Euronews reports. Common Sense Media founder and CEO James P. Steyer is quoted in the launch statement saying, "AI is reshaping childhood and adolescence, yet we are making critical decisions about children's futures without the evidence we need to ensure it's safe and in their interest," per Euronews. Editorial analysis: Industry observers will scrutinise whether a consumer-style rating model can be adapted to rapidly updating, context-sensitive AI services used by children.
Quick Read+3

Showing 24 articles

See the full news feed